[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"item-843":3,"related-843":62},{"id":4,"title":5,"url":6,"summary":7,"summary_zh":8,"content":9,"source_name":10,"source_url":8,"published_at":11,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":15,"score_detail":16,"sources":23,"tags":25,"primary_topic":31,"secondary_topics":37,"classification":38,"search_phrases":57,"slug":60,"view_count":52,"doi":8,"paper":8,"created_at":61},843,"基于EfficientNet与FPN多尺度特征融合的田间害虫智能检测(原标题:Enhancing Pest Detection in Agriculture: A Multi-Scale Feature Fusion Approach with YOLOv3)","https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591","华北大学Zhang等基于EfficientNet与FPN多尺度特征融合构建田间害虫智能检测框架:EfficientNet作为轻量级注意力嵌入骨干网络提取层次特征,集成多尺度检测与分层FPN融合以提升微小、不显眼的害虫识别性能。在37种田间作物常见害虫上的实验表明,所提模型在IoU阈值0.5下的平均精度(mAP@0.5)达到98.89%、平均识别错误率1.57%、平均推理时间仅0.048秒\u002F图像,在检测精度和实时性能间取得平衡,为田间害虫监测提供轻量、可靠、自动化的监测解决方案,促进数据驱动、精准的害虫管理,推动可持续绿色精准农业。",null,"## 1. Introduction\n\nSustainable production of staple cereal (e.g., wheat and maize) and oilseed crops (e.g., soybean and canola) underpins global agricultural stability and catalyzes the advancement of precision and ecologically sustainable farming systems [[1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B1-agronomy-16-01591),[2](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B2-agronomy-16-01591)]. However, pest infestation remains the predominant biotic constraint on crop productivity, accounting for global yield losses ranging from 10% to 28%, with even more severe losses reported in developing countries [[3](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B3-agronomy-16-01591)]. Additionally, pests are indirectly responsible for major losses in few plants, such as cankers and shoot blights in nuts, since many insects serve as a vector for pathogens [[4](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B4-agronomy-16-01591)]. Beyond significant yield reduction, pest infestation often compels farmers to increase pesticide application, hinders the development of green and sustainable agriculture [[5](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B5-agronomy-16-01591),[6](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B6-agronomy-16-01591)]. These adverse effects further disrupt routine field management, reduce crop productivity, and hinder the implementation of precision agriculture, forming a core bottleneck limiting the stability and sustainability of agricultural production systems. With the global transition toward digital and precision agriculture, mitigating pest-induced hazards is essential for advancing agronomic development and optimizing crop cultivation and management [[7](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B7-agronomy-16-01591)]. Traditional pest monitoring relies heavily on manual field surveys, which are characterized by low operational efficiency, delayed feedback loops, and limited spatial coverage [[8](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B8-agronomy-16-01591)]. In the absence of robust, reliable monitoring systems, large-scale pest outbreaks can emerge readily—compromising agroecosystem integrity and significantly reducing crop productivity [[9](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B9-agronomy-16-01591)].\n\nIn recent years, advances in deep learning and computer vision have built a new intelligent agricultural risk management paradigm [[10](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B10-agronomy-16-01591)]. Vision-based pest detection has become a mainstream research direction, outperforming manual surveys in field practicality [[11](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B11-agronomy-16-01591)]. Convolutional neural network (CNN)-based pest identification methods automatically extract hierarchical visual features to enable fast, high-accuracy pest classification [[12](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B12-agronomy-16-01591),[13](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B13-agronomy-16-01591)]. The hierarchical features extracted by the VGG network are shown in [Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f001)a. Compared with traditional methods, deep learning detection techniques have superior performance in real-time capability, scalability and accuracy, relying on standardized unified frameworks. As illustrated in [Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f001)b, mainstream object detectors adopt a four-stage structure: input, backbone, neck and head ([Figure 1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f001)b). These detectors are broadly categorized into single-stage (dense prediction) and two-stage (sparse prediction) paradigms. Each paradigm further subdivides into anchor-based and anchor-free approaches. Representative dense-prediction models include the anchor-based YOLO and RetinaNet, as well as the anchor-free CenterNet and FCOS. In contrast, sparse-prediction models encompass the anchor-based Faster R-CNN and the anchor-free RepPoints. Among these algorithms, the YOLO series models have attracted considerable attention due to their excellent balance between detection speed and accuracy [[14](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B14-agronomy-16-01591),[15](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B15-agronomy-16-01591)].\n\n**Figure 1.** Schematic diagram of convolutional neural network feature extraction and object detection structure. (**a**) Schematic diagram of feature extraction in convolutional neural networks; (**b**) Schematic diagram of the target detection structure.\n\nAlthough YOLO-based architectures have demonstrated robust performance in agricultural pest monitoring, several critical methodological limitations persist in practical field-scale crop pest detection tasks. First, the original Darknet-53 backbone of YOLOv3 lacks attention modules, failing to capture subtle features of tiny pests in complex field backgrounds [[16](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B16-agronomy-16-01591),[17](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B17-agronomy-16-01591)], leading to misclassification of similar pest species and frequent missed detections [[18](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B18-agronomy-16-01591)]. Second, the primitive simple upsampling feature fusion module of original YOLOv3 fails to construct complete top-down multi-scale feature transmission, weakening the model’s ability to distinguish small-size pests against cluttered leaf [[19](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B19-agronomy-16-01591),[20](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B20-agronomy-16-01591)]. Existing feature fusion frameworks rarely construct bidirectional feature circulation, restricting miniature pest recognition [[21](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B21-agronomy-16-01591),[22](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B22-agronomy-16-01591)]. Third, most detection models adopt generic anchor boxes pre-trained on public natural image datasets. The size distribution of these anchor boxes may not match miniature field pests, leading to frequent missed localization of small pest individuals. Dataset-adaptive anchor clustering effectively alleviates this issue [[23](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B23-agronomy-16-01591),[24](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B24-agronomy-16-01591)]. Fourth, existing YOLO variants struggle to balance detection precision and computational overhead, forming an unaddressed trade-off dilemma [[25](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B25-agronomy-16-01591)]. Deep heavy backbones deliver favorable recognition accuracy but incur excessive inference latency, which cannot be deployed on low-power portable edge devices for field real-time monitoring [[26](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B26-agronomy-16-01591)]. While lightweight YOLO-tiny models cut computational cost by simplifying network layers, yet the compressed feature extraction capacity drastically reduces detection accuracy for small and overlapping pests [[27](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B27-agronomy-16-01591)]. Furthermore, few datasets have been specially annotated for wheat, corn, soybean and canola pests, restricting the targeted optimization and benchmark evaluation of detection algorithms for dominant field crops [[28](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B28-agronomy-16-01591),[29](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B29-agronomy-16-01591)]. To address the above shortcomings, this study proposes an optimized YOLOv3-based intelligent pest detection framework with three targeted optimizations: replacing Darknet-53 with lightweight attention-equipped EfficientNet as the feature extraction backbone, introducing a complete Feature Pyramid Network (FPN) for comprehensive multi-scale feature fusion, and comparing the COCO anchor boxes with the anchor boxes that are adaptively generated from the dataset through K-means clustering, thereby selecting the scale features that are more suitable for field pests. Additionally, a manually annotated multi-class pest dataset covering four major field crops is constructed to validate the performance of the optimized detection model. The proposed model demonstrates enhanced pest detection performance and delivers a computationally efficient, lightweight detection solution, thereby providing technical support for intelligent and precision monitoring of crop pests in agricultural fields.\n\n## 2. Material and Methods\n\n### 2.1. Dataset\n\nThe experimental image data in this paper is sourced from the Anhui Academy of Agricultural Sciences [[30](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B30-agronomy-16-01591),[31](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B31-agronomy-16-01591)], covering 4285 images of 37 types of pests found in several common field crops, including corn, soybean, wheat, and canola. All images were captured by the use of digital cameras (e.g., Canon, Nikon, Tokyo, Japan, and mobile devices). Among them, 3162 images (73.8% of the entire dataset) contained only a single pest individual; the remaining 1123 images (26.2% of the total) contained two or more overlapping pest instances. Furthermore, Following the standard pixel-area-based threshold definition used in YOLO-based detection tasks, we classified annotated pest objects into three size categories and computed their respective proportions: small targets (pixel area \u003C 32 2) constitute 47.2%, medium targets (32 2 ≤ pixel area \u003C 96 2) account for 39.6%, and large targets (pixel area ≥ 96 2) represent only 13.2%. The predominance of small-scale pests further underscores the necessity of our multi-scale feature fusion architecture. To eliminate the potential negative effects of illumination variability, all sample images were preprocessed with uniform illumination settings in crop field situations, as was done by Wen and Guyer (2012) [[32](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B32-agronomy-16-01591)]. The various types of pest examples included in the dataset are shown in [Figure 2](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f002), and the names and quantities of each type of pest are listed in [Table 1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#table_body_display_agronomy-16-01591-t001).\n\n**Figure 2.** Sample images from the pest dataset, along with their corresponding identifiers, which match those listed in [Table 1](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#table_body_display_agronomy-16-01591-t001).\n\n**Table 1.** Pest infestation dataset: image names and quantities.\n\n### 2.2. Data Processing\n\nTo train the model, it is necessary to label the pests in the dataset first. LabelImg is a tool for marking the position and name of the target in the image in deep learning. It is written in Python 3.12 and uses PyQt as its graphical interface. During the annotation process, the annotations are saved as XML files in the PASCAL VOC format. The images of the labeled pests and the generated XML files after labeling are shown in [Figure 3](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f003).\n\n**Figure 3.** Sample images and annotated pest files.\n\nSecondly, the dataset constitutes the foundational input for training deep neural network models, and its quality critically determines the performance and generalizability of the resulting models. Prior to model construction, it is essential to design an input data processing strategy aligned with the chosen network architecture commonly referred to as dataset preprocessing. This preprocessing stage typically comprises two key components: image rescaling and image augmentation.\n\nAll pest images in this dataset have an original resolution of 200 × 200 pixels. However, the YOLOv3 network model expects input images of size 416 × 416 pixels by default. Thus, all images are first resized to this dimension. This upsampling operation inevitably introduces interpolation artifacts, which smooth fine high-frequency spatial gradients and attenuate subtle edge-texture features, potentially degrading the model’s capability in feature extraction and precise localization of tiny pest targets. To mitigate the adverse effects of such gradient distortion, we implement a suite of coordinated optimization strategies: (i) the multi-scale feature fusion architecture inherent to YOLO-series networks preserves delicate local gradient information through shallow feature branches, specifically enhancing sensitivity to small objects; (ii) dataset-adaptive for anchor box generation ensures anchor dimensions closely match the actual pixel scales of miniature pests, thereby reducing localization bias arising from blurred target boundaries; and (iii) L2 regularization and gradient accumulation configured via the hyperparameters jointly stabilize gradient updates that would otherwise fluctuate due to resizing-induced artifacts.\n\nA unified baseline augmentation pipeline is implemented identically across all four models to guarantee fair cross-architecture comparison. All augmentation operations are applied on-the-fly during training rather than generating a static offline expanded dataset; the raw dataset retains a fixed total of 4285 original images, with unique random distortions generated dynamically for every training iteration. The shared augmentation suite includes standardized geometric and photometric transformations, with full parameter constraints defined as follows: (1) Isotropic random scaling: Each image and its corresponding bounding boxes are scaled synchronously with a randomly sampled scaling factor ranging from 0.5 to 1.5, which avoids extreme shrinkage or magnification that would obscure pest morphological features; (2) Random translation: Horizontal and vertical translation offsets are restricted within ±15% of the total image width and height to prevent partial cropping of pest bodies. (3) Horizontal flipping and color jittering: Random horizontal mirroring is performed with a 50% probability, while mild brightness, contrast and saturation jitter are applied to enrich photometric diversity without distorting pest texture identities. Meanwhile, to prevent morphological distortion that could lead to inter-species pest misidentification, strict amplitude constraints are imposed on all geometric transformations. Concurrently, shallow feature extraction branches retain fine-grained texture gradients, while COCO anchor boxes are more suitable for the natural size distribution of the target pests in this research dataset. These measures collectively enhancing the model’s ability to discriminate visually similar pest individuals, even under augmentation-induced distortions. Since all YOLO variants are bounding-box–based object detectors, ground-truth bounding box coordinates are transformed in strict synchrony with each augmentation operation, thereby preserving consistent spatial alignment between input images and their corresponding annotation labels.\n\nBeyond the shared baseline preprocessing, YOLOv4 and YOLOv4-tiny incorporate an exclusive additional data augmentation technique named Mosaic augmentation, a refined variant derived from CutMix [[33](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B33-agronomy-16-01591)], while Mosaic augmentation is not activated during the training of YOLOv3 and YOLOv3-tiny. The detailed operating procedure of Mosaic augmentation is as follows. Firstly, the Mosaic data augmentation technique combines four different input images (each with its corresponding bounding box) into a single composite image. Following this integration, a new image and its corresponding bounding boxes are generated and subsequently fed into the neural network for training and optimization. Specifically, each of the four constituent images is first independently subjected to a series of geometric and photometric transformations, including horizontal flipping, isotropic scaling, and color-space jittering. The transformed images are then spatially arranged in the top-left, bottom-left, bottom-right, and top-right quadrants of the output canvas, respectively. Finally, fixed subregions from each quadrant are extracted via matrix indexing and seamlessly stitched together to form the final mosaic image, whose merged bounding boxes are fed into the network for training and optimization.\n\n### 2.3. The Framework Structure of the YOLO Model\n\nYOLOv3 is developed based on YOLOv1 and YOLOv2. As shown in [Figure 4](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#fig_body_display_agronomy-16-01591-f004), it is a fully convolutional network with a total of 252 layers, using Darknet53 as the backbone feature extraction network [[34](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B34-agronomy-16-01591)]. To enhance the detection ability for small objects, the network adopts the methods of feature map upsampling and multi-scale feature fusion, conducting detection on feature maps of multiple scales. The input is the image of pests. The backbone feature extraction network adopts the 52 convolutional layers of Darknet53 after removing the last fully connected layer, which consists of 5 groups of ResNet residual blocks. Each group of residual blocks is repeated 1, 2, 8, 8, and 4 times, respectively. During each repetition process, a 1 × 1 convolution is executed first, followed by a 3 × 3 convolution, and finally the filter is halved and then restored. YOLO takes the entire insect damage image as the input of the network. After the backbone network extracts the feature map of a certain size (S × S), the entire insect damage image is divided into S × S grids. Each grid predicts three bounding boxes. If the center of a certain insect damage target falls within a certain grid, then this grid is responsible for predicting this insect damage target. The output layer outputs the position of the bounding box and the category to which the insect damage target belongs. Each bounding box predicts four positions (x, y, w, h) and one confidence, totaling five values. Each grid also predicts one category information, with a total of C categories (there are 37 types of insect damage in the dataset of this article). For one anchor box, the total number of predicted channels equals 5 + 37 = 42; since each grid predicts three boxes, the total channel count per grid is 3 × 42 = 126, and ultimately the output tensor of each detection branch takes the shape of S × S × 126, where S × S stands for the spatial grid partition of the feature map, channels 1–4 correspond to the bounding box regression parameters (x, y, w, h) of the first anchor box, channel 5 represents the objectness confidence score of the first anchor box, channels 6–42 denote the class probability distribution of the first anchor box, and this identical channel layout is repeated sequentially for the second and third anchor boxes within every grid cell.\n\n**Figure 4.** Schematic diagram of the YOLOv3 structure.\n\n**Figure 4.** Schematic diagram of the YOLOv3 structure.\n\n[![Image 1: Agronomy 16 01591 g004](https:\u002F\u002Fwww.mdpi.com\u002Fagronomy\u002Fagronomy-16-01591\u002Farticle_deploy\u002Fhtml\u002Fimages\u002Fagronomy-16-01591-g004.png)](https:\u002F\u002Fwww.mdpi.com\u002Fagronomy\u002Fagronomy-16-01591\u002Farticle_deploy\u002Fhtml\u002Fimages\u002Fagronomy-16-01591-g004.png)\n\nYOLOv3-tiny is a compressed version of the YOLOv3 network, which has removed the residual layers and only uses two YOLO output layers of different scales. The network has a total of 23 layers, including 13 convolutional layers, 6 pooling layers, 1 routing layer, 1 upsampling layer, and 2 output layers. Except for the convolutional layer before the YOLO output layer, which uses a linear activation function, each convolutional layer is followed by a BN layer, and each convolutional layer is followed by a convolutional layer with a leaky rectified linear unit (Leaky ReLU) activation function [[35](https:\u002F\u002Fwww.mdpi.com\u002F2073-4395\u002F16\u002F16\u002F1591#B35-agronomy-16-01591)]. The output dimensions of the two output layers are","MDPI Agronomy 16(16):1591 2026年8月18日","2026-08-18T00:00:00Z","论文",10,false,76,{"impact":17,"substance":18,"depth":17,"authority":19,"freshness":20,"relevant":21,"comment":22},18,20,12,8,1,"论文提出多尺度特征融合的害虫检测方法，方法新颖，数据详实，对智慧农业有实质贡献。",[24],{"name":10,"url":6},[26,27,28,29,30],"精准农业","害虫检测","EfficientNet","FPN多尺度融合","MDPI Agronomy",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},"A7","forage-cropping","饲草与耕作制度","\u002Ftopic\u002Ffield-crops\u002Fforage-cropping","A",[],{"status":39,"locked":14,"version":40,"method":41,"evidence":42},"classified","2026-10-22.2","rules",{"candidates":43,"agricultural_evidence":47},[44,51],{"code":45,"score":46,"supported":47,"title":48,"abstract":49,"tags":50},"I1",4,true,[],[26],[26],{"code":32,"score":52,"supported":47,"title":53,"abstract":54,"tags":56},3,[],[55],"间作",[],[58,59],"FPN多尺度融合 害虫检测 精准农业 MDPI Agronomy","FPN多尺度融合 害虫检测","FPN多尺度融合害虫检测精准农业MDPIAgronomy-843","2026-08-22T00:06:00.849334Z",{"total":63,"page":21,"page_size":63,"items":64},6,[65,124,229,318,367,416],{"id":66,"title":67,"url":68,"summary":69,"summary_zh":8,"content":70,"source_name":71,"source_url":8,"published_at":72,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":73,"score_detail":74,"sources":78,"tags":80,"primary_topic":85,"secondary_topics":86,"classification":87,"search_phrases":118,"slug":121,"view_count":122,"doi":8,"paper":8,"created_at":123},202,"Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops(像素级可解释压力指数:田间作物病害严重程度量化的语义分割框架)","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585","植物病害每年造成全球农业产量损失20%-40%、经济损失超2200亿美元。文章提出统一深度学习流水线,集成语义分割、基于回归的严重程度估计和病害分类,根据感染叶片面积比例将胁迫严重程度分为四档。在包含1641样本、六个类别的苹果树叶病害分割数据集上,采用MobileNetV2的U-Net表现最佳,像素准确率98.20%、mIoU 0.70、检测准确率99.41%、单张图像处理时间仅14.7毫秒,适合实时田间应用。计算出的严重程度指数与专家标注高度相关(r=0.968, R²=0.937),为自动化作物监测和决策支持提供可靠方案。[预印本]","[Skip to main content](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585#content)\n\n![Image 1](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbase\u002F1.0.1\u002Fimages\u002Ficons\u002Fsmileybones-small.svg)arXiv is now an independent nonprofit![Learn more](https:\u002F\u002Finfo.arxiv.org\u002Fabout)×\n\n[](https:\u002F\u002Farxiv.org\u002FIgnoreMe)[![Image 2: archive](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbase\u002F1.0.1\u002Fimages\u002Farxiv-logo-primary-light.svg)](https:\u002F\u002Farxiv.org\u002F)[Search](https:\u002F\u002Farxiv.org\u002Fsearch)[Submit](https:\u002F\u002Farxiv.org\u002Fuser\u002Fcreate)[Donate](https:\u002F\u002Finfo.arxiv.org\u002Fabout\u002Fdonate.html)[Log in](https:\u002F\u002Farxiv.org\u002Flogin)\n\nSearch arXiv \n\n Press Enter to search · [Advanced search](https:\u002F\u002Farxiv.org\u002Fsearch\u002Fadvanced)\n\n# Computer Science > Computer Vision and Pattern Recognition\n\n**arXiv:2607.06585** (cs) \n\n [Submitted on 5 Jul 2026]\n\n# Title:Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops\n\nAuthors:[Raunak Kumar](https:\u002F\u002Farxiv.org\u002Fsearch\u002Fcs?searchtype=author&query=Kumar,+R), [Soumyashree Kar](https:\u002F\u002Farxiv.org\u002Fsearch\u002Fcs?searchtype=author&query=Kar,+S)\n\nView a PDF of the paper titled Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops, by Raunak Kumar and Soumyashree Kar\n\n[View PDF](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2607.06585)[HTML (experimental)](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2607.06585v1)\n> Abstract:Plant diseases, resulting from both biotic and abiotic stresses, cause an estimated 20-40% loss in global agricultural yield annually, resulting in economic damages exceeding USD 220 billion. Accurate and scalable stress quantification is essential for precision agriculture, yet traditional manual assessments are labour-intensive and subjective. This paper proposes a unified deep learning pipeline integrating semantic segmentation, regression-based severity estimation, and disease classification. Stress severity is categorised into four levels (Low to Very High) based on the proportion of infected leaf area. Experiments on the Apple Tree Leaf Disease Segmentation dataset (1,641 samples, six classes) evaluate four models: U-Net (MobileNetV2), SegFormer, FCN, and PSPNet. U-Net with MobileNetV2 achieves the best performance with 98.20% pixel accuracy, 0.70 mIoU, and 99.41% detection accuracy at 14.7 ms per image, making it suitable for real-time use. SegFormer performs competitively (mIoU 0.66), while FCN and PSPNet show lower spatial accuracy (approximately 0.49 mIoU). The computed severity index strongly correlates with expert annotations (r = 0.968, R^2 = 0.937), demonstrating the system's reliability for automated crop monitoring and decision support.\n\nComments:26 pages, 15 figures, 5 tables\nSubjects:Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)\nCite as:[arXiv:2607.06585](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585) [cs.CV]\n(or [arXiv:2607.06585v1](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585v1) [cs.CV] for this version)\n[https:\u002F\u002Fdoi.org\u002F10.48550\u002FarXiv.2607.06585](https:\u002F\u002Fdoi.org\u002F10.48550\u002FarXiv.2607.06585)\n\nFocus to learn more\n\n arXiv-issued DOI via DataCite\n\n## Submission history\n\n From: Raunak Kumar [[view email](https:\u002F\u002Farxiv.org\u002Fshow-email\u002F62311445\u002F2607.06585)] \n\n**[v1]** Sun, 5 Jul 2026 06:04:20 UTC (2,899 KB)\n\n[](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)Full-text links:\n## Access Paper:\n\n View a PDF of the paper titled Pixel-Precise Explainable Stress Indexing: A Semantic Segmentation Framework for Disease Severity Quantification in Field Crops, by Raunak Kumar and Soumyashree Kar\n\n*   [View PDF](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2607.06585)\n*   [HTML (experimental)](https:\u002F\u002Farxiv.org\u002Fhtml\u002F2607.06585v1)\n*   [TeX Source](https:\u002F\u002Farxiv.org\u002Fsrc\u002F2607.06585)\n\n[view license](http:\u002F\u002Farxiv.org\u002Flicenses\u002Fnonexclusive-distrib\u002F1.0\u002F \"Rights to this article\")\n\n### Current browse context:\n\ncs.CV\n\n[\u003Cprev](https:\u002F\u002Farxiv.org\u002Fprevnext?id=2607.06585&function=prev&context=cs.CV \"previous in cs.CV (accesskey p)\") | [next>](https:\u002F\u002Farxiv.org\u002Fprevnext?id=2607.06585&function=next&context=cs.CV \"next in cs.CV (accesskey n)\")\n\n[new](https:\u002F\u002Farxiv.org\u002Flist\u002Fcs.CV\u002Fnew) | [recent](https:\u002F\u002Farxiv.org\u002Flist\u002Fcs.CV\u002Frecent) | [2026-07](https:\u002F\u002Farxiv.org\u002Flist\u002Fcs.CV\u002F2026-07)\n\n Change to browse by: \n\n[cs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585?context=cs)\n\n[cs.LG](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585?context=cs.LG)\n\n### References & Citations\n\n*   [NASA ADS](https:\u002F\u002Fui.adsabs.harvard.edu\u002Fabs\u002FarXiv:2607.06585)\n*   [Google Scholar](https:\u002F\u002Fscholar.google.com\u002Fscholar_lookup?arxiv_id=2607.06585)\n*   [Semantic Scholar](https:\u002F\u002Fapi.semanticscholar.org\u002FarXiv:2607.06585)\n\nexport BibTeX citation Loading...\n\n## BibTeX formatted citation\n\n×\n\nData provided by: [](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)\n\n### Bookmark\n\n[![Image 3: BibSonomy](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbrowse\u002F0.3.4\u002Fimages\u002Ficons\u002Fsocial\u002Fbibsonomy.png)](http:\u002F\u002Fwww.bibsonomy.org\u002FBibtexHandler?requTask=upload&url=https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585&description=Pixel-Precise%20Explainable%20Stress%20Indexing:%20A%20Semantic%20Segmentation%20Framework%20for%20Disease%20Severity%20Quantification%20in%20Field%20Crops \"Bookmark on BibSonomy\")[![Image 4: Reddit](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbrowse\u002F0.3.4\u002Fimages\u002Ficons\u002Fsocial\u002Freddit.png)](https:\u002F\u002Freddit.com\u002Fsubmit?url=https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585&title=Pixel-Precise%20Explainable%20Stress%20Indexing:%20A%20Semantic%20Segmentation%20Framework%20for%20Disease%20Severity%20Quantification%20in%20Field%20Crops \"Bookmark on Reddit\")\n\nBibliographic Tools \n\n# Bibliographic and Citation Tools\n\n- [x] Bibliographic Explorer Toggle \n\nBibliographic Explorer _([What is the Explorer?](https:\u002F\u002Finfo.arxiv.org\u002Flabs\u002Fshowcase.html#arxiv-bibliographic-explorer))_\n\n- [x] Connected Papers Toggle \n\nConnected Papers _([What is Connected Papers?](https:\u002F\u002Fwww.connectedpapers.com\u002Fabout))_\n\n- [x] Litmaps Toggle \n\nLitmaps _([What is Litmaps?](https:\u002F\u002Fwww.litmaps.co\u002F))_\n\n- [x] scite.ai Toggle \n\nscite Smart Citations _([What are Smart Citations?](https:\u002F\u002Fwww.scite.ai\u002F))_\n\nCode, Data, Media \n\n# Code, Data and Media Associated with this Article\n\n- [x] alphaXiv Toggle \n\nalphaXiv _([What is alphaXiv?](https:\u002F\u002Falphaxiv.org\u002F))_\n\n- [x] Links to Code Toggle \n\nCatalyzeX Code Finder for Papers _([What is CatalyzeX?](https:\u002F\u002Fwww.catalyzex.com\u002F))_\n\n- [x] DagsHub Toggle \n\nDagsHub _([What is DagsHub?](https:\u002F\u002Fdagshub.com\u002F))_\n\n- [x] GotitPub Toggle \n\nGotit.pub _([What is GotitPub?](http:\u002F\u002Fgotit.pub\u002Ffaq))_\n\n- [x] Huggingface Toggle \n\nHugging Face _([What is Huggingface?](https:\u002F\u002Fhuggingface.co\u002Fhuggingface))_\n\n- [x] ScienceCast Toggle \n\nScienceCast _([What is ScienceCast?](https:\u002F\u002Fsciencecast.org\u002Fwelcome))_\n\nDemos \n\n# Demos\n\n- [x] Replicate Toggle \n\nReplicate _([What is Replicate?](https:\u002F\u002Freplicate.com\u002Fdocs\u002Farxiv\u002Fabout))_\n\n- [x] Spaces Toggle \n\nHugging Face Spaces _([What is Spaces?](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Fhub\u002Fspaces))_\n\n- [x] Spaces Toggle \n\nTXYZ.AI _([What is TXYZ.AI?](https:\u002F\u002Ftxyz.ai\u002F))_\n\nRelated Papers \n\n# Recommenders and Search Tools\n\n- [x] Link to Influence Flower \n\nInfluence Flower _([What are Influence Flowers?](https:\u002F\u002Finfluencemap.cmlab.dev\u002F))_\n\n- [x] Core recommender toggle \n\nCORE Recommender _([What is CORE?](https:\u002F\u002Fcore.ac.uk\u002Fservices\u002Frecommender))_\n\n*   [Author](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)\n*   [Venue](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)\n*   [Institution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)\n*   [Topic](https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.06585)\n\n About arXivLabs  \n\n# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [**Learn more about arXivLabs**](https:\u002F\u002Finfo.arxiv.org\u002Flabs\u002Findex.html).\n\n[Which authors of this paper are endorsers?](https:\u002F\u002Farxiv.org\u002Fauth\u002Fshow-endorsers\u002F2607.06585) | [Disable MathJax](javascript:setMathjaxCookie()) ([What is MathJax?](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Fmathjax.html)) \n\n We gratefully acknowledge support from our **major funders**, [**member institutions**](https:\u002F\u002Finfo.arxiv.org\u002Fabout\u002Fourmembers.html), , and all contributors. \n\n[About](https:\u002F\u002Finfo.arxiv.org\u002Fabout)·[Help](https:\u002F\u002Finfo.arxiv.org\u002Fhelp)·[Contact](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Fcontact.html)·[Subscribe](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Fsubscribe)·[Copyright](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Flicense\u002Findex.html)·[Privacy](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Fpolicies\u002Fprivacy_policy.html)·[Accessibility](https:\u002F\u002Finfo.arxiv.org\u002Fhelp\u002Fweb_accessibility.html)·[Operational Status (opens in new tab)](https:\u002F\u002Fstatus.arxiv.org\u002F)\n\nMajor funding support from\n\n[![Image 5: Simons Foundation](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbase\u002F1.0.1\u002Fimages\u002Ffunders\u002Fsimons-foundation.png)](https:\u002F\u002Fwww.simonsfoundation.org\u002F)[![Image 6: Simons Foundation International](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbase\u002F1.0.1\u002Fimages\u002Ffunders\u002Fsimons-foundation-international.png)](https:\u002F\u002Fwww.sfi.org.bm\u002F)[![Image 7: Schmidt Sciences](https:\u002F\u002Farxiv.org\u002Fstatic\u002Fbase\u002F1.0.1\u002Fimages\u002Ffunders\u002Fschmidt-sciences.png)](https:\u002F\u002Fwww.schmidtsciences.org\u002F)","arXiv:2607.06585v1 [cs.CV] 2026-07-05","2026-07-05T00:00:00Z",72,{"impact":17,"substance":75,"depth":17,"authority":19,"freshness":76,"relevant":21,"comment":77},22,2,"论文提出基于语义分割的作物病害严重度量化框架，方法新颖且性能优异，对智慧农业有参考价值，但时效性较低。",[79],{"name":71,"url":68},[81,82,26,83,84],"可解释AI","语义分割","病害严重程度","实时监测",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[],{"status":39,"locked":14,"version":40,"method":41,"evidence":88},{"candidates":89,"agricultural_evidence":47},[90,94,101,108,114],{"code":32,"score":20,"supported":47,"title":91,"abstract":92,"tags":93},[55],[],[],{"code":95,"score":63,"supported":47,"title":96,"abstract":97,"tags":100},"B2",[],[98,99],"果树","苹果",[],{"code":102,"score":103,"supported":47,"title":104,"abstract":105,"tags":107},"I2",5,[],[106],"深度学习",[81,82],{"code":109,"score":52,"supported":47,"title":110,"abstract":111,"tags":113},"J3",[],[112],"全球农业",[],{"code":45,"score":21,"supported":14,"title":115,"abstract":116,"tags":117},[],[],[26],[119,120],"病害严重程度 实时监测 精准农业 语义分割","病害严重程度 实时监测","病害严重程度实时监测精准农业语义分割-202",9,"2026-08-04T01:15:47.534597Z",{"id":125,"title":126,"url":127,"summary":128,"summary_zh":129,"content":8,"source_name":130,"source_url":127,"published_at":131,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":132,"score_detail":133,"sources":137,"tags":139,"primary_topic":145,"secondary_topics":146,"classification":159,"search_phrases":194,"slug":197,"view_count":198,"doi":199,"paper":200,"created_at":228},5573,"Improving alfalfa nitrogen estimation with canopy structural priors from DSSAT–PROSAIL and UAV hyperspectral data","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.eja.2026.128375","Accurate estimation of nitrogen (N) concentration in alfalfa ( Medicago sativa L .) supports forage quality assessment and harvest scheduling. Empirical remote sensing models often overfit because of confounding effects of three-dimensional canopy structure. Incorporating structural parameters can stabilize biochemical inversions, but concurrent in situ morphological measurements are difficult to acquire at scale. We developed the Morphology-Aware Latent Embedding Network (MALENet), a hybrid framework integrating physiological and morphological mechanisms with hyperspectral data. The Decision Support System for Agrotechnology Transfer (DSSAT), coupled with the PROSAIL radiative transfer model, generated a structurally rich synthetic hyperspectral prior. A convolutional neural network (CNN) with a Squeeze-and-Excitation (SE) attention mechanism, was pre-trained on these data to distill ecophysiological knowledge into a dense latent representation. This mechanistically derived embedding was then fused with empirical hyperspectral measurements collected under diverse field conditions. Evaluation using multi-year field quadrats in the Hexi Corridor showed improved predictive performance ( R² = 0.71, RMSE = 0.30%) relative to conventional machine learning and deep learning models (LASSO-ANN: R² = 0.61; CNN: R² = 0.66). Feature attribution indicated that MALENet used near-infrared multiple scattering (58.3% attribution) as a structural proxy to calibrate nitrogen concentration retrieval against the confounding effects of canopy architecture. Despite constraints imposed by sample size and PROSAIL's 1D canopy assumption, these findings demonstrate the potential of physically guided deep learning to use structural information for agricultural trait retrieval in precision forage management.","准确估算紫花苜蓿（Medicago sativa L.）的氮浓度有助于饲草品质评估和收获期安排。经验性遥感模型常因三维冠层结构的混杂效应而出现过拟合。引入结构参数可以稳定生化反演，但同步开展原位形态学测量难以大规模获取。我们开发了形态感知潜在嵌入网络（Morphology-Aware Latent Embedding Network，MALENet），这是一个将生理与形态机制同高光谱数据相结合的混合框架。农业技术转移决策支持系统（Decision Support System for Agrotechnology Transfer，DSSAT）与PROSAIL辐射传输模型耦合，生成了结构信息丰富的合成高光谱先验数据。在此基础上，预训练了一个带有挤压—激励（Squeeze-and-Excitation，SE）注意力机制的卷积神经网络（convolutional neural network，CNN），将生态生理学知识提炼为稠密潜在表征。随后，将这一机制性导出的嵌入与在不同田间条件下采集的经验性高光谱测量数据相融合。利用河西走廊多年田间样方进行的评估显示，相较于传统机器学习和深度学习模型（LASSO-ANN：R² = 0.61；CNN：R² = 0.66），预测性能有所提升（R² = 0.71，RMSE = 0.30%）。特征归因表明，MALENet利用近红外多次散射（归因占比58.3%）作为结构代理，以校准氮浓度反演，抵御冠层结构的混杂效应。尽管受样本量和PROSAIL一维冠层假设的限制，这些发现表明，物理引导的深度学习有潜力利用结构信息，在精准饲草管理中进行农业性状反演。","European Journal of Agronomy","2026-10-21T00:00:00Z",80,{"impact":134,"substance":75,"depth":17,"authority":135,"freshness":13,"relevant":21,"comment":136},16,14,"该研究将作物模型与辐射传输模型耦合生成结构先验，并结合无人机高光谱数据反演苜蓿氮素，方法新颖、田间验证扎实，对精准饲草管理有参考价值，值得进入每日精选。",[138],{"name":130,"url":127},[140,141,142,143,144],"智慧农业","农业人工智能","农业遥感","氮素监测","紫花苜蓿",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[147,153],{"code":148,"slug":149,"name":150,"path":151,"parent_code":152},"I3","remote-sensing","遥感与农情监测","\u002Ftopic\u002Fdigital-agriculture\u002Fremote-sensing","I",{"code":154,"slug":155,"name":156,"path":157,"parent_code":158},"C4","feed","饲料与饲草","\u002Ftopic\u002Flivestock\u002Ffeed","C",{"status":39,"locked":14,"version":40,"method":41,"evidence":160},{"candidates":161,"agricultural_evidence":47},[162,169,176,182,190],{"code":148,"score":17,"supported":47,"title":163,"abstract":165,"tags":168},[164],"hyperspectral",[166,167,164],"遥感","remote sensing",[142],{"code":32,"score":170,"supported":47,"title":171,"abstract":173,"tags":175},15,[172],"alfalfa",[174,172],"苜蓿",[144],{"code":154,"score":13,"supported":47,"title":177,"abstract":178,"tags":181},[],[179,174,180],"饲草","forage",[144],{"code":102,"score":13,"supported":47,"title":183,"abstract":184,"tags":189},[],[106,185,186,187,188],"机器学习","machine learning","deep learning","decision support",[141],{"code":45,"score":21,"supported":14,"title":191,"abstract":192,"tags":193},[],[],[140],[195,196],"DSSAT PROSAIL 苜蓿 氮素","无人机 高光谱 苜蓿","DSSATPROSAIL苜蓿氮素-5573",0,"10.1016\u002Fj.eja.2026.128375",{"doi":199,"openalex_id":201,"authors":202,"venue":130,"cited_by_count":198,"oa_url":127,"card":221,"direction":225,"ingested_from":227},"W7223154576",[203,205,207,209,211,214,216,219],{"name":204,"orcid":8},"Shuai Fu",{"name":206,"orcid":8},"Yunhao Li",{"name":208,"orcid":8},"Chunli Miao",{"name":210,"orcid":8},"Jie Liu",{"name":212,"orcid":213},"Tiangang Liang","https:\u002F\u002Forcid.org\u002F0000-0001-5599-6587",{"name":215,"orcid":8},"Yongjia Hu",{"name":217,"orcid":218},"Qisheng Feng","https:\u002F\u002Forcid.org\u002F0009-0000-1391-9591",{"name":220,"orcid":8},"吴彩霞",{"tldr":222,"method":223,"finding":224,"direction":225,"opportunity":226},"构建MALENet融合DSSAT-PROSAIL结构先验与无人机高光谱，提升苜蓿氮浓度估算精度。","DSSAT耦合PROSAIL生成合成高光谱先验，预训练CNN-SE网络，再融合田","MALENet精度优于传统机器学习和深度学习模型，R²达0.71，近红外多次散射作为结构代理贡献58","农业遥感与作物表型","可探索将作物模型结构先验迁移至其他牧草或作物生化参数反演，并缓解PROSAIL一维冠层假设局限。","openalex","2026-10-22T11:20:39.622844Z",{"id":230,"title":231,"url":232,"summary":233,"summary_zh":234,"content":8,"source_name":235,"source_url":232,"published_at":236,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":237,"score_detail":238,"sources":240,"tags":242,"primary_topic":245,"secondary_topics":246,"classification":248,"search_phrases":270,"slug":273,"view_count":198,"doi":274,"paper":275,"created_at":317},5567,"Capturing Spatial Heterogeneity Using UAV-Derived Multispectral Phenotypes and Random Regression Models Improves Genetic Evaluation in Alfalfa Breeding Trials","https:\u002F\u002Fdoi.org\u002F10.64898\u002F2026.10.14.759249","Spatial heterogeneity is a major source of environmental noise in field trials, reducing genetic evaluation accuracy. Conventional spatial correction methods use field layout and plot proximity but often fail to capture in-season dynamics or leverage temporal data from high-throughput phenotyping. We hypothesized that longitudinal UAV-derived multispectral measurements could characterize field-scale environmental heterogeneity and improve genetic evaluation in breeding trials. Multispectral imagery was collected throughout the growing season across eight alfalfa trials in New York and New Mexico (20222023). Vegetation indices were evaluated for heritability and association with biomass yield using univariate and multi-trait models. Longitudinal measurements were modeled with random regression (RR) to estimate plot-specific permanent environmental effects (PEE), incorporated into genomic mixed models within a two-stage framework. Model performance was compared with conventional autoregressive (AR) and baseline (no correction) models using metrics of consistency and accuracy of estimated genetic effects across replicates. Vegetation indices showed moderate to high heritability, with GNDVI consistently outperforming NDVI and NDRE. Both univariate and multi-trait analyses revealed substantial spatial variability, particularly in New Mexico trials. RR-derived PEE captured field-scale spatial patterns and improved genetic signal recovery relative to baseline and AR models, with the largest gains under strong spatial heterogeneity and comparable performance where variability was limited. The RR framework improved reproducibility of genetic effect estimates across independent field subsets without compromising agreement with full-trial evaluations. These findings show that integrating UAV data via random regression enhances genetic signal recovery and selection accuracy, with broader potential for precision agriculture and biosensor-driven crop management.","空间异质性是田间试验中环境噪声的主要来源，会降低遗传评估的准确性。传统空间校正方法利用田间布局和小区邻近性，但往往无法捕捉生育期内的动态变化，也未能充分利用高通量表型分析产生的时间序列数据。我们假设，基于无人机（UAV）的纵向多光谱测量能够表征田块尺度的环境异质性，并改善育种试验中的遗传评估。在2022—2023年纽约州和新墨西哥州的8个紫花苜蓿试验中，于整个生长季收集了多光谱影像。采用单性状和多性状模型评估了植被指数的遗传力及其与生物量产量的关联。纵向测量数据采用随机回归（RR）建模，以估计小区特异性永久环境效应（PEE），并在两阶段框架中将其纳入基因组混合模型。利用重复间估计遗传效应的一致性和准确性指标，将模型性能与传统的自回归（AR）模型和基线（无校正）模型进行比较。植被指数表现出中到高的遗传力，其中GNDVI始终优于NDVI和NDRE。单性状和多性状分析均揭示了显著的空间变异，尤其在新墨西哥州的试验中更为明显。RR衍生的PEE捕捉到了田块尺度的空间格局，相对于基线模型和AR模型改善了遗传信号的恢复，在强空间异质性条件下增益最大，而在变异有限时性能相当。RR框架提高了遗传效应估计在独立田间子集间的可重复性，同时未损害与全试验评估的一致性。这些结果表明，通过随机回归整合UAV数据可增强遗传信号恢复和选择准确性，在精准农业和生物传感器驱动的作物管理方面具有更广泛的潜力。","bioRxiv (Cold Spring Harbor Laboratory)","2026-10-20T00:00:00Z",79,{"impact":17,"substance":75,"depth":17,"authority":19,"freshness":122,"relevant":21,"comment":239},"该研究利用无人机多光谱表型与随机回归模型校正田间空间异质性，显著提升苜蓿育种遗传评估精度，方法新颖、数据覆盖两州八组试验，对智慧育种有实质参考价值。",[241],{"name":235,"url":232},[140,243,244,166,144],"无人机","育种",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[247],{"code":148,"slug":149,"name":150,"path":151,"parent_code":152},{"status":39,"locked":14,"version":40,"method":41,"evidence":249},{"candidates":250,"agricultural_evidence":47},[251,255,260,266],{"code":32,"score":170,"supported":47,"title":252,"abstract":253,"tags":254},[172],[174,172],[144],{"code":148,"score":19,"supported":47,"title":256,"abstract":258,"tags":259},[257],"multispectral",[257],[166],{"code":45,"score":261,"supported":47,"title":262,"abstract":263,"tags":265},7,[],[26,264],"precision agriculture",[140],{"code":154,"score":46,"supported":47,"title":267,"abstract":268,"tags":269},[],[174],[144],[271,272],"无人机 多光谱 苜蓿 育种","随机回归 空间异质性 遗传评估","无人机多光谱苜蓿育种-5567","10.64898\u002F2026.10.14.759249",{"doi":274,"openalex_id":276,"authors":277,"venue":235,"cited_by_count":198,"oa_url":232,"card":312,"direction":225,"ingested_from":227},"W7223083966",[278,281,284,287,290,292,294,297,300,303,306,309],{"name":279,"orcid":280},"Dinesh Ghimire","https:\u002F\u002Forcid.org\u002F0009-0008-2281-1396",{"name":282,"orcid":283},"Ranjita Thapa","https:\u002F\u002Forcid.org\u002F0000-0001-6116-7611",{"name":285,"orcid":286},"Karl H. Kunze","https:\u002F\u002Forcid.org\u002F0000-0003-4548-1808",{"name":288,"orcid":289},"Felipe Sabadin","https:\u002F\u002Forcid.org\u002F0000-0003-2937-1465",{"name":291,"orcid":8},"Julie Hansen",{"name":293,"orcid":8},"Christopher Pierce",{"name":295,"orcid":296},"Liam Wickes-Do","https:\u002F\u002Forcid.org\u002F0009-0003-2927-4590",{"name":298,"orcid":299},"Nicholas Santantonio","https:\u002F\u002Forcid.org\u002F0000-0002-4351-4023",{"name":301,"orcid":302},"Ian M. Ray","https:\u002F\u002Forcid.org\u002F0000-0003-2056-536X",{"name":304,"orcid":305},"Michael A. Gore","https:\u002F\u002Forcid.org\u002F0000-0001-6896-8024",{"name":307,"orcid":308},"Virginia Mae Moore","https:\u002F\u002Forcid.org\u002F0000-0001-7888-3366",{"name":310,"orcid":311},"Kelly R. Robbins","https:\u002F\u002Forcid.org\u002F0000-0001-9522-9585",{"tldr":313,"method":314,"finding":315,"direction":225,"opportunity":316},"用无人机多光谱时序表型和随机回归模型校正田间空间异质性，提升苜蓿育种遗传评估精度。","八项苜蓿试验的无人机多光谱植被指数，随机回归估计小区永久环境效应并嵌入基因组混合","随机回归提取的环境效应能捕捉田块尺度空间格局，在强异质性下显著提升遗传信号恢复与选择准确性。","可将无人机时序表型与随机回归框架推广到其他作物育种试验，并探索多源遥感与基因组选择的联合建模。","2026-10-22T11:20:32.655394Z",{"id":319,"title":320,"url":321,"summary":322,"summary_zh":8,"content":323,"source_name":324,"source_url":8,"published_at":236,"category":325,"cover_url":8,"hotness":13,"is_selected":14,"score":326,"score_detail":327,"sources":331,"tags":333,"primary_topic":339,"secondary_topics":340,"classification":347,"search_phrases":362,"slug":365,"view_count":198,"doi":8,"paper":8,"created_at":366},5426,"浙江农科院在景宁发布丘陵梯田稻薯轮作全程机械化与共富模式","http:\u002F\u002Fwww.came.net.cn\u002FsecondLevelPage\u002Finfo\u002F31\u002F227412","浙江省农业科学院联合浙江省自然资源集团、景宁畲族自治县政府在景宁举办丘陵山区多田套合工程示范暨丘陵梯田全程机械化稻薯轮作现场会。项目由省农科院农业装备研究所牵头，核心示范 200 亩、辐射 1000 亩。研发的小土豆移栽机作业效率较人工提升 8 倍以上，「无人机吊运+轨道运输+地面农机」体系下单次吊运 50 公斤以上，山地运输效率提升 10 倍以上、单位运输成本降低 60% 以上；病毒学与生物技术研究所攻关的茎尖脱毒组培苗移栽成活率不低于 90%，脱毒种薯出苗率不低于 98%。","近日，由浙江省农业科学院、浙江省自然资源集团有限公司、景宁畲族自治县人民政府联合主办的丘陵山区多田套合工程示范暨丘陵梯田全程机械化稻薯轮作生产技术集成与共富模式示范现场会在景宁畲族自治县顺利举办。活动系统呈现了省农科院联合省自然资源集团、浙江默梵绿芯种业有限公司等多方主体在景宁探索构建的丘陵山区稻薯轮作全程机械化与产业链共富新模式。\n\n省农科院党委委员、副院长冯冠胜致辞并宣布现场会开幕，他指出，当前浙江省深入推进共同富裕示范区建设，山区是补齐发展短板、激发内生潜力的关键区域。省农科院围绕多田套合、稻薯轮作全程机械化与产业链共富开展协同攻关、集成示范、落地推广，取得了阶段性成果。下一步，将紧贴山区农业产业发展实际需求，充分发挥多良集成优势，持续深化技术攻关、成果熟化、示范推广和落地服务，全力助力景宁农业产业提质增效，系统总结可复制、可推广的丘陵山地产业共富模式，为全国丘陵山区农业高质量发展贡献农科力量。\n\n该项目由省农科院农业装备研究所牵头，联合作物与核技术利用研究所、病毒学与生物技术研究所等共同实施。针对丘陵山区稻薯轮作茬口衔接难、专用适配农机短缺、农产品销路不稳等难题，项目集成土地、科技、产业要素，探索山区产业共富路径：依托省自然资源集团国企资金资源优势，实施“多田套合”土地整治工程，将零散“巴掌田”改造为连片宜机梯田；依托省农科院农机农艺融合科研优势，筛选“薯稻薯”菜用土豆与水稻品种，攻克茎尖脱毒组培与移栽调控茬口技术，研发小土豆移栽装备，构建“无人机吊运+轨道运输+地面农机”协同作业体系；依托浙江默梵绿芯种业有限公司市场品牌优势，推行订单保底收购、品质溢价激励机制，保障农户稳定增收。该模式核心示范200亩、辐射1000亩。\n\n现场展示的丘陵山区稻薯轮作全程机械化生产技术模式，集成精细化整地、机械化种植、轻简化管理、低空吊运+轨道运输、机械化收获等环节。研发适配山地的小土豆移栽机，作业效率较人工提升8倍以上；“无人机吊运+轨道运输+地面农机”协同作业模式中，无人机单次可吊运50公斤以上农机、农资、秧盘，山地运输效率较人工提升10倍以上，单位运输成本降低60%以上。现场还演示了智能灌溉、机械化耕整地、精准植保及收获等成套装备，并展示巡检机器狗、作物无人机监测等智能装备与技术。\n\n在农艺技术方面，省农科院作物与核技术利用研究所开展适宜“薯稻薯”模式的水稻品种筛选，评价耐寒性、早熟性、抗病性及机械化适配性；病毒学与生物技术研究所开展菜用土豆品种筛选与茎尖脱毒组培快繁技术攻关，组培苗移栽成活率不低于90%，可调控茬口5至10天；脱毒种薯出苗率不低于98%，较自留种增产20%至40%。通过品种搭配、播期衔接与密度调控，构建“薯—稻—薯”一年多熟轮作体系，提升复种指数与种薯质量。\n\n产业链运营方面，项目打造“企业+合作社+农户”铁三角模式，统一品种标准、种植技术、品质管控、产品收购全流程标准化管理，推行订单保底收购、品质溢价激励机制，并依托共富工坊、农旅融合等新业态拓宽村集体与农户增收渠道。活动还举行丘陵山地机械化应用合作签约、稻薯轮作种植推广联合体签约及农机农艺融合综合示范基地揭牌仪式。\n\n本次现场会展示的模式实现三大转变，即生产方式从“人力依赖”转向“机械智控”、种植模式从“零散粗放”转向“标准高效”、产业发展从“单一种植”转向“全链共富”，通过稳技术、稳标准、稳销路、稳收益，真正实现山地农业产业提质、农户增收、集体增效的良性共富新格局。\n\n发布时间：2026-10-20提交人：浙江省畜牧技术推广与种畜禽监测总站（浙江省农业机械试验鉴定推广总站） 董玉玲责任编辑：农业农村部农机化总站 信息宣传处","中国农业机械化信息网 2026-10-20","报道",81,{"impact":75,"substance":328,"depth":134,"authority":329,"freshness":122,"relevant":21,"comment":330},21,13,"省级科研机构牵头、多主体协同的丘陵山区稻薯轮作全程机械化与共富模式示范，含具体装备效率与农艺数据，信息增量足，具行业推广参考价值。",[332],{"name":324,"url":321},[334,335,336,337,338],"农机农艺融合","丘陵山区","全程机械化","共富模式","稻薯轮作",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[341],{"code":342,"slug":343,"name":344,"path":345,"parent_code":346},"F5","machinery","农机装备与机械化","\u002Ftopic\u002Finputs-equipment\u002Fmachinery","F",{"status":39,"locked":14,"version":40,"method":41,"evidence":348},{"candidates":349,"agricultural_evidence":47},[350,356],{"code":342,"score":134,"supported":47,"title":351,"abstract":353,"tags":355},[352],"机械化",[354,352],"农机",[336,334],{"code":32,"score":357,"supported":47,"title":358,"abstract":360,"tags":361},11,[359],"轮作",[359],[],[363,364],"浙江农科院 景宁 稻薯轮作","丘陵梯田 全程机械化","浙江农科院景宁稻薯轮作-5426","2026-10-21T00:10:00.508230Z",{"id":368,"title":369,"url":370,"summary":371,"summary_zh":8,"content":372,"source_name":373,"source_url":8,"published_at":374,"category":325,"cover_url":8,"hotness":13,"is_selected":14,"score":375,"score_detail":376,"sources":379,"tags":381,"primary_topic":384,"secondary_topics":385,"classification":386,"search_phrases":411,"slug":414,"view_count":198,"doi":8,"paper":8,"created_at":415},5211,"省农科院助力景宁破解丘陵山区种田难:多田套合+稻薯轮作新模式","https:\u002F\u002Fwww.jingning.gov.cn\u002Fcol\u002Fcol1376112\u002Fart\u002F2026\u002Fart_de8faa5b86ef48648e94bfbd06c59191.html","2026年10月15日,丘陵山区多田套合工程示范暨丘陵梯田全程机械化稻薯轮作生产技术集成与共富模式示范现场会在浙江景宁举行,浙江省农科院与浙江省自然资源集团合作,通过政企研协同推广\"多田套合+稻薯轮作\"让传统山地面貌焕新。","10月15日，丘陵山区多田套合工程示范暨丘陵梯田全程机械化稻薯轮作生产技术集成与共富模式示范现场会在景宁东坑镇儒章村举行。来自省农业科学院、省自然资源集团以及市县农业农村和自然资源系统的专家、企业代表、种植大户齐聚现场，聚焦丘陵山区土地整治、稻薯轮作技术推广、山地农机装备应用等，共同探索山区农业现代化共富新模式。\n\n开幕仪式上，浙江丽水市自然资源发展有限公司与省农科院农业装备研究所签订丘陵山地机械化应用合作协议；浙江默梵绿芯种业有限公司与景宁种植大户代表签订稻薯轮作种植推广联合体合作协议；现场为“农机农艺融合综合示范基地”揭牌。\n\n仪式结束后，与会人员前往东坑镇儒章村多田套合核心示范区开展现场观摩。连片整治后的梯田错落铺开，一改过去碎片化、零散化的面貌。现场，多田套合土地整治工程成果、优质稻薯新品种、轻简化种植模式逐一展示。\n\n无人旋耕机、巡检机器狗、微耕机、无人机吊运小农机、水稻割晒打捆机、小土豆移栽机等一批适配山区小地块的农机装备轮番作业演示。轨道自动化运输设备穿梭梯田之间，破解山地物资转运难题，过去梯田“人工作业、肩挑背扛”的传统劳作场景被现代化农机替代，让“农机上山”从设想变为现实。\n\n本次现场会由浙江省农业科学院、浙江省自然资源集团、景宁畲族自治县人民政府联合主办，聚焦丘陵山区现实短板，集成土地整治、优良品种、山地农机、稻薯轮作、共富经营整套模式。通过政企研协同发力，把科研院所实验室里的技术成果，落地到畲乡的丘陵梯田，在保障粮食生产的基础上，挖掘土特产产业链价值，拓宽山区群众增收渠道，为全省丘陵山区农业现代化提供景宁实践样本。","景宁县政府(2026-10-16)","2026-10-16T00:00:00Z",66,{"impact":17,"substance":377,"depth":135,"authority":122,"freshness":20,"relevant":21,"comment":378},17,"省级科研机构与地方政府共建丘陵山区全程机械化稻薯轮作示范，含实质技术集成与多方合作，具行业参考价值。",[380],{"name":373,"url":370},[140,382,335,383,338],"农机装备","土地整治",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[],{"status":39,"locked":14,"version":40,"method":41,"evidence":387},{"candidates":388,"agricultural_evidence":47},[389,393,397,402,407],{"code":32,"score":357,"supported":47,"title":390,"abstract":391,"tags":392},[359],[359],[],{"code":342,"score":46,"supported":47,"title":394,"abstract":395,"tags":396},[],[352],[382],{"code":398,"score":21,"supported":14,"title":399,"abstract":400,"tags":401},"G2",[],[],[383],{"code":403,"score":21,"supported":14,"title":404,"abstract":405,"tags":406},"H1",[],[],[383],{"code":45,"score":21,"supported":14,"title":408,"abstract":409,"tags":410},[],[],[140],[412,413],"景宁 稻薯轮作 丘陵山区","浙江省农科院 山地农机","景宁稻薯轮作丘陵山区-5211","2026-10-18T00:05:23.005537Z",{"id":417,"title":418,"url":419,"summary":420,"summary_zh":8,"content":8,"source_name":421,"source_url":419,"published_at":422,"category":12,"cover_url":8,"hotness":13,"is_selected":14,"score":423,"score_detail":424,"sources":426,"tags":428,"primary_topic":433,"secondary_topics":434,"classification":445,"search_phrases":462,"slug":465,"view_count":198,"doi":466,"paper":467,"created_at":493},4652,"Linking land use advantage and temporal stability for row configuration design in maize\u002Fsoybean intercropping under water-limited dryland conditions in northeastern China","https:\u002F\u002Fdoi.org\u002F10.1016\u002Fj.agsy.2026.105005","Linking land use advantage and temporal stability for row configuration design in maize\u002Fsoybean intercropping under water-limited dryland conditions in northeastern China。Agricultural Systems","Agricultural Systems","2026-10-10T00:00:00Z",74,{"impact":19,"substance":328,"depth":17,"authority":135,"freshness":122,"relevant":21,"comment":425},"核心期刊论文，方法新颖、结论可靠，但属细分领域研究，产业影响有限，可作为专业精选。",[427],{"name":421,"url":419},[55,429,430,431,432],"玉米大豆带状复合种植","东北旱作","水分利用","作物行配置",{"code":32,"slug":33,"name":34,"path":35,"parent_code":36},[435,440],{"code":436,"slug":437,"name":438,"path":439,"parent_code":36},"A3","maize","玉米","\u002Ftopic\u002Ffield-crops\u002Fmaize",{"code":441,"slug":442,"name":443,"path":444,"parent_code":36},"A4","soybean-oilseeds","大豆与油料","\u002Ftopic\u002Ffield-crops\u002Fsoybean-oilseeds",{"status":39,"locked":14,"version":40,"method":41,"evidence":446},{"candidates":447,"agricultural_evidence":47},[448,453,457],{"code":32,"score":19,"supported":47,"title":449,"abstract":451,"tags":452},[450],"intercropping",[450],[55],{"code":436,"score":357,"supported":47,"title":454,"abstract":455,"tags":456},[437],[437],[],{"code":441,"score":357,"supported":47,"title":458,"abstract":460,"tags":461},[459],"soybean",[459],[],[463,464],"玉米大豆间作 行配置 东北","旱地 间作 产量稳定性","玉米大豆间作行配置东北-4652","10.1016\u002Fj.agsy.2026.105005",{"doi":466,"openalex_id":468,"authors":469,"venue":421,"cited_by_count":198,"oa_url":419,"card":8,"direction":8,"ingested_from":227},"W7221334648",[470,473,476,478,480,482,485,487,489,491],{"name":471,"orcid":472},"Yue Zhang","https:\u002F\u002Forcid.org\u002F0000-0003-0506-409X",{"name":474,"orcid":475},"Xiaoyi Wang","https:\u002F\u002Forcid.org\u002F0000-0003-3843-2434",{"name":477,"orcid":8},"Shixiang Xi",{"name":479,"orcid":8},"Hongjun Du",{"name":481,"orcid":8},"Chen Feng",{"name":483,"orcid":484},"Zhe Zhang","https:\u002F\u002Forcid.org\u002F0000-0002-4579-7709",{"name":486,"orcid":8},"Erjing Guo",{"name":488,"orcid":8},"Wei Bai",{"name":490,"orcid":8},"Lizhen Zhang",{"name":492,"orcid":8},"Zhanxiang Sun","2026-10-11T23:30:03.973971Z"]