Rice bacterial leaf blight disease resistance screening method based on unmanned aerial vehicle and deep learning

Abstract

The invention relates to the technical field of agricultural disease detection and crop disease resistance screening, in particular to a rice bacterial leaf blight disease resistance screening method based on an unmanned aerial vehicle and deep learning, and adopts the technical scheme that an optimized YOLOv11-OBB model is composed of a C3k2FC module, an SPPFLSKA module, a SlimNeck module and a LiteHead module, is responsible for extracting features in an image and positioning bacterial leaf blight spots, and is used for screening the disease resistance of the rice bacterial leaf blight; bacterial leaf blight can be efficiently and accurately detected; a lightweight deep learning model is used, efficient operation on a resource-limited unmanned aerial vehicle platform can be realized, and the equipment and calculation cost is reduced; through the cooperation of a plurality of optimization modules, the detection precision of bacterial leaf blight spots is remarkably improved, and a stable detection effect can be kept; the operation process is simplified, and full automation of bacterial blight disease resistance screening is realized; therefore, low-cost, efficient, automatic and high-precision bacterial blight disease resistance screening is realized by utilizing the lightweight and optimized YOLOv11-OBB model and combining module design with high calculation efficiency, and the method is suitable for large-scale field application.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural disease detection and crop disease resistance screening, and particularly relates to a rice bacterial leaf blight disease resistance screening method based on a UAV and deep learning. BACKGROUND

[0002] Rice is one of the most important food crops in the world, and bacterial leaf blight is one of the main diseases affecting rice yield and quality. Traditional bacterial leaf blight disease resistance screening methods rely on manual observation and recording, which have problems such as high labor intensity, strong subjectivity, and poor data consistency. In addition, existing bacterial leaf blight disease resistance screening methods, although they have improved screening efficiency, still face limitations such as high equipment costs, high computing resource requirements, and complex operations. Specifically, existing UAV platforms rely on high-precision sensors, resulting in high equipment costs; traditional deep learning models require strong computing resources and cannot efficiently run on low-cost, computing resource-limited UAV platforms; existing technologies usually rely on professional personnel for data analysis and operation, and lack efficient automated screening systems.

[0003] In view of this, we propose a rice bacterial leaf blight disease resistance screening method based on a UAV and deep learning to solve the existing problems. SUMMARY

[0004] The purpose of the present application is to provide a rice bacterial leaf blight disease resistance screening method based on a UAV and deep learning to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application provides the following technical solution: a rice bacterial leaf blight disease resistance screening method based on a UAV and deep learning, the operation steps of which include: S1, UAV image acquisition: the UAV flies at a predetermined flight route and height, and takes pictures of bacterial leaf blight lesion images in the rice field; S2, bacterial leaf blight lesion detection and positioning: after inputting the image data, the YOLOv11-OBB model is used to detect the bacterial leaf blight lesions; the improved YOLOv11-OBB model is composed of a C3k2FC module, a SPPF_LSKA module, a SlimNeck module, and a LiteHead module, which are responsible for extracting features in the image and positioning the bacterial leaf blight lesions; S3, target tracking and multi-target identification: a Kalman filter is used to track the bacterial leaf blight lesions in each frame of image for multi-target tracking, avoiding the situation of image repetition of the same target during flight; S4, disease incidence calculation and disease resistance classification: the disease incidence of rice is calculated according to the number of detected lesions, and the rice is classified into different disease resistance grades using the threshold value of the incidence, so as to screen the rice varieties according to the disease resistance; S5, screening result output and display: the final screening result is output and presented to the user in the form of a graphical interface or report, helping the breeding researcher to make a selection of disease-resistant varieties; In S2, the overall structure of the improved YOLOv11-OBB model includes a backbone, a neck, and a detection head. The improvement includes: in the backbone network, the C3k2FC module integrates PConv and CGLU into the C3k2 structure; in the backbone network, the SPPF_LSKA module includes LSKA pooling; in the neck, the SlimNeck module uses GSConv and VoVGSCSP to optimize feature fusion while minimizing FLOP; in the detection head, the LiteHead module uses separate batch normalization and dynamic anchor boxes; in addition, LAMP pruning implements adaptive global pruning to compress model parameters.

[0006] Further, for the C3k2FC module, two 1x1 point convolutions in the FasterNet block are replaced with the CGLU module, forming an improved FC structure; at the same time, PConv is integrated into the deep convolution section of the C3k2 bottleneck part, finally forming an improved C3k2FC module.

[0007] Further, for the SPPF_LSKA module, the structure first extracts multi-resolution features through multiple maximum pooling branches, and then applies the LSKA module to attention-weighted spliced features; the horizontal and vertical branches of LSKA process spatial information, and the depth separable convolution further optimizes channel mixing; finally, the fused features are output through a 1x1 convolution.

[0008] Further, for the SlimNeck module, the GSConv unit introduces a Shuffle mechanism to fuse channel-related data obtained from SC with spatial output from DSC.

[0009] Further, for the SlimNeck module, the VoVGSCSP unit divides the input matrix channel into two parts; one part is preprocessed through convolution and extracts features through a series of GS bottleneck units, while the other part is used as a residual path and only performs a single convolution transformation.

[0010] Further, the core components of the LiteHead detection head include shared convolution layers, separate BN modules, and dynamic anchor box generators.

[0011] Further, LAMP determines the pruning threshold by calculating the adaptive importance score of each layer.

[0012] Further, the core components of the GSConv unit include Conv, DWConv, Concat, and Shuffle operations.

[0013] Further, the VoVGSCSP unit utilizes CSP framework enhancement features reuse, minimizes data loss through a one-time aggregation method, and integrates batch normalization and SiLU activation functions to maintain feature consistency.

[0014] Further, shared convolution layers are used for multi-task reuse; separate BN modules apply BN independently for each detection branch; and a dynamic anchor generator calculates anchor parameters in real time based on feature map statistics.

[0015] Compared with the prior art, the beneficial effects of the present application are: The optimized YOLOv11-OBB model provided by the present application can efficiently and accurately detect white leaf blight spots, is suitable for automatic screening of large-scale rice fields, uses a lightweight deep learning model that can efficiently run on a resource-constrained unmanned aerial vehicle platform, reduces equipment and computing costs, significantly improves the detection accuracy of white leaf blight spots through the cooperation of multiple optimization modules, especially in complex environments, leaf occlusion and light changes, etc., can maintain stable detection effect, simplifies the operation process, realizes the full automation of white leaf blight resistance screening, and provides a more efficient solution for rice breeding; therefore, by using the lightweight and optimized YOLOv11-OBB model, combined with the module design of high computing efficiency, low-cost, efficient, automatic, and high-precision white leaf blight resistance screening is realized, which is suitable for large-scale field application. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 Fig. 1 is a structural schematic diagram of the white leaf blight detection model of the present application; Figure 2 Fig. 2 is a structural schematic diagram of the C3k2FC module of the present application; Figure 3 Fig. 3 is a structural schematic diagram of the SPPF_LSKA module of the present application; Figure 4 Fig. 4 is a structural schematic diagram of the SlimNeck module of the present application; Figure 5 Fig. 5 is a schematic diagram of the LAMP process of the present application. DETAILED DESCRIPTION

[0017] The technical solutions of the present application are further described below in conjunction with the drawings and specific embodiments. Embodiment one

[0018] The rice white leaf blight resistance screening method based on unmanned aerial vehicles and deep learning includes the following operation steps: S1, unmanned aerial vehicle image acquisition: the unmanned aerial vehicle flies at a predetermined flight route and height, and takes pictures of the white leaf blight spot in the rice field; the flight height is set to 0.6-1.5 meters, and the image resolution is 1920x1080 pixels; the images captured by the unmanned aerial vehicle are transmitted to the ground computing device through wireless or wired mode; light unmanned aerial vehicles such as DJI Mini 2 or DJI Mini 4 Pro are used for image acquisition of white leaf blight infection in rice field, to ensure full coverage, data consistency and accuracy; S2, white leaf blight spot detection and positioning: after inputting image data, the YOLOv11-OBB model is used for white leaf blight spot detection; the improved YOLOv11-OBB model is composed of C3k2FC module, SPPF_LSKA module, SlimNeck module and LiteHead module, which is responsible for extracting features in the image and positioning white leaf blight spots; based on YOLOv11-OBB model, after optimization of C3k2FC, SPPF_LSKA, SlimNeck and LiteHead modules, the detection accuracy and calculation efficiency of white leaf blight spots are significantly improved; S3, target tracking and multi-target identification: Kalman filter is used for multi-target tracking of white leaf blight spots in each frame of image, to avoid occlusion and repeated appearance of the same target in the image during flight; the position of the lesion is predicted by Kalman filter to reduce the error and repetition in detection; S4, disease incidence calculation and disease resistance classification: the disease incidence of rice is calculated according to the number of detected lesions; the rice is divided into different disease resistance grades (L1, L2, L3, L4) by using the threshold value of incidence, so as to select rice varieties with strong disease resistance; disease resistance screening is automatically completed by algorithm, avoiding human error; S5, output and display of screening results: the final screening results are output and presented to the user in the form of graphical interface or report, helping breeders to make disease resistance variety selection; rice varieties with strong white leaf blight resistance are selected according to disease resistance grade for use in breeding research.

[0019] YOLOv11 is an advanced object detection model, which provides multiple versions, including standard YOLOv11, YOLOv11-OBB, YOLOv11-Seg for segmentation tasks and YOLOv11-Pose for pose estimation. In addition, YOLOv11 has multiple size variants: n, s, l and x, each of which is optimized for different calculation efficiency and performance.

[0020] Since the UAV captures images from above, the target usually presents at different angles. In addition to this, white leaf blight usually appears as elongated lesions. Considering these factors, the YOLOv11-OBB model was chosen. To optimize the model efficiency and reduce the computational complexity, the smallest n model was chosen as the basis for improvement. As shown in Figure 1 the overall structure of the improved model includes three main parts: Backbone, Neck, and Head.

[0021] The improvement includes: in the backbone network, the C3k2FC module integrates PConv and CGLU into the C3k2 structure to reduce redundancy and improve efficiency; in the backbone network, the SPPF_LSKA module combines LSKA pooling to enhance multi-scale feature aggregation and reduce noise; in the neck, the SlimNeck module uses GSConv and VoVGSCSP to optimize feature fusion while minimizing FLOP; in the detection head, the LiteHead module uses separate batch normalization (BN) and dynamic anchor boxes to better handle rotation; in addition, LAMP pruning implements adaptive global pruning to compress model parameters while maintaining accuracy.

[0022] The C3k2 module in YOLOv11-OBB is based on the Cross Stage Partial (CSP) structure, which divides the input feature map into two parallel branches. This design enhances feature reuse and gradient flow by repeating bottleneck units, supports flexible layer and channel configuration, and balances model depth and computational efficiency. However, this module has redundant convolution operations, increasing the computational cost and making deployment on resource-constrained UAV embedded systems complex.

[0023] To solve these problems, the FasterNet block's partial convolution (PConv) and TransNext CGLU module are integrated into the C3k2 module to enhance the backbone and neck network. This improvement aims to achieve lightweight design, reduce computational redundancy, and minimize parameter size, making it more suitable for deployment on resource-constrained UAV embedded systems. The CGLU module is an evolved version of the GLU Gated Linear Unit, which expands local spatial awareness by inserting a 3x3 depth convolution before the activation function. This module consists of two linear transformation paths: one path applies a 3x3 depth convolution, followed by a gated activation function to enhance local feature capture, and the other path generates a baseline output. These two paths are combined by element-wise multiplication to adjust feature weights, and finally generate the final output through linear projection and residual connection with the input.

[0024] The FasterNet block uses a reverse residual structure to simplify the network, first using a 1x1 point convolution for channel expansion and feature compression. Then the PConv module is inserted, which applies a 3x3 standard convolution for spatial feature mixing to a quarter of the input channels, while the remaining channels are passed through an identity mapping to reduce full-channel computation. Next, the feature distribution and nonlinearity are maintained through a normalization layer and a ReLU activation function, and finally the channels are restored and integrated into the output through another 1x1 point convolution.

[0025] As shown in Figure 2 The present application replaces the two 1x1 point convolutions in the FasterNet block with a CGLU module, forming an improved FC structure. At the same time, the PConv is integrated into the deep convolution section of the C3k2 bottleneck part, finally forming an improved C3k2FC module.

[0026] This structure optimizes feature extraction through partial channel convolution and gating mechanism, reducing floating point operations and memory access. The FC block with PConv module extracts multi-scale features from the input image while maintaining low computational overhead. Combined with the CGLU module, this design supports dynamic feature weight allocation. The improved C3k2FC module integrates these components to capture detailed and global information, supporting rice white leaf blight detection and positioning under the perspective of unmanned aerial vehicles, suitable for deployment on edge devices.

[0027] The SPPF module in YOLOv11 is based on the spatial pyramid pooling (SPPF) structure, which aggregates multi-scale features by applying different size max-pooling operations in parallel. Then these features are fused through channel concatenation and convolution to extract global context. This module aims to reduce information loss and enhance adaptability to different scales of targets, supporting efficient feature representation. However, the pooling operation of this module has limitations, leading to low efficiency in feature aggregation in complex scenarios, and may introduce background noise. This is particularly challenging when dealing with rice field images captured by unmanned aerial vehicles, as there are different lighting, leaf occlusion and resolution fluctuations in the image, which can cause fine-grained details to be ignored or irrelevant features to be amplified, leading to false detection and performance degradation.

[0028] To solve these problems, the present application integrates a large separable convolution attention mechanism (LSKA) into the SPPF module, thereby improving multi-scale feature extraction and local attention capability. The LSKA module is based on separable convolution, which decomposes large kernel convolution into 1D convolution in horizontal and vertical directions to simulate a large receptive field attention mechanism. This method combines depth separable convolution to reduce the number of parameters.

[0029] As shown in Figure 3As shown, the SPPF_LSKA module integrates the LSKA mechanism into the pooling layer of traditional SPPF. This structure first extracts multi-resolution features through multiple max-pooling branches, then applies the LSKA module to perform attention weighting on the concatenated features. The horizontal and vertical branches of LSKA process spatial information, and depthwise separable convolutions further optimize channel blending. Finally, the fused features are output through a 1×1 convolution to enhance global and local semantic representations.

[0030] This application replaces the original SPPF module in YOLOv11-OBB with an improved SPPF_LSKA module, and optimizes the integration of multi-scale features through an attention mechanism to support the detection and localization of rice bacterial blight from the perspective of UAVs.

[0031] The standard convolutional (SC) component in YOLOv11-OBB processes multi-channel data through parallel multi-core operations. While this facilitates deep interaction between channels, it also leads to parameter inflation and a significant increase in floating-point operations (FLOPs), limiting real-time performance, especially on resource-constrained UAV embedded systems. On the other hand, efficient architectures like MobileNet and ShuffleNet alleviate the computational burden by using depthwise separable convolutions (DSC). However, DSC isolates channel data during computation, which greatly reduces the effectiveness of feature integration and extraction. This limitation is particularly challenging for detecting rice bacterial blight spots from a UAV perspective, as variations in lighting, leaf occlusion, and resolution fluctuations in images can lead to small spots being missed or inaccurately located.

[0032] To accelerate computation without compromising detection accuracy, this application integrates the GSConv composite convolutional unit into the neck structure of YOLOv11-OBB. The GSConv unit introduces a shuffle mechanism to fuse channel correlation data obtained from the SC with the spatial output from the DSC. Compared to using DSC alone, GSConv reduces the computational burden while preserving potential channel correlations, thereby optimizing accuracy and response time, making it ideal for crop disease detection on edge hardware.

[0033] The core components of the GSConv unit include Conv, DWConv, Concat, and Shuffle operations. Its structure is as follows: the input feature matrix contains C1 channels; depthwise separable convolution (DSC) is applied to one half of the channels to capture spatial details, while standard convolution (SC) is applied to the other half for channel fusion; the outputs from both sides are concatenated along the channel axes; a Shuffle operation is applied to the concatenated result to promote random channel rearrangement and enhance data interaction; the final output matrix contains C2 channels.

[0034] VoVGSCSP is an iterative bottleneck combination of GSConv that divides the input matrix channels into two parts. One part undergoes convolutional preprocessing and extracts features through a series of GS bottleneck units, while the other part serves as a residual path, undergoing only a single convolutional transformation. This unit leverages the CSP framework to enhance feature reuse, minimizes data loss through a one-time aggregation method, and integrates batch normalization and SiLU activation functions to maintain feature consistency.

[0035] like Figure 4 As shown, this application reconstructs the neck framework of YOLOv11-OBB using GSConv and VoVGSCSP, forming a SlimNeck structure. These modifications reduce the computational burden of the model, improve inference speed and data processing capabilities, and enhance the integration of multi-resolution features. This optimization scheme achieves an effective balance between speed and accuracy.

[0036] The YOLOv11-OBB detection head uses a shared convolution mechanism to process multi-scale feature outputs for bounding box prediction. While this approach promotes parameter reuse, the use of uniform batch normalization (BN) leads to inaccurate moving averages when there are significant statistical differences between feature layers, thus affecting training stability and generalization performance. Furthermore, the static anchor box design struggles to adapt to the rotation and scale variations of white leaf blight spots, especially during drone photography, where challenges such as leaf occlusion, resolution fluctuations, and changes in lighting conditions often result in localization errors or missed small targets, thus limiting overall detection performance.

[0037] To address these issues, this application proposes the LiteHead detection head. This optimized design aims to improve computational efficiency, reduce parameter size and operational burden, making the system more suitable for resource-constrained UAV embedded systems. LiteHead avoids Batch Normalization (BN) correlation problems by separately handling statistical differences between features at different levels, ensuring independent moving averages and thus preventing the accumulation of biases in shared parameters. Furthermore, a dynamic anchor frame generator dynamically adjusts the scale, angle, and aspect ratio of the anchor frames based on input features, thereby enhancing the alignment capability for rotating targets.

[0038] The core components of the LiteHead detection head include: a shared convolutional layer for multi-task reuse, reducing redundant parameters; a separate Batch Normalization (BN) module, which applies BN independently to each detection branch to optimize feature distribution and accelerate convergence; and a dynamic anchor box generator, which calculates anchor box parameters in real time based on feature map statistics, supporting the prediction of rotated bounding boxes (OBB). Its workflow is as follows: multi-scale feature maps are processed through shared convolutions to extract bounding box regression and classification scores; the output of each layer is normalized by separate BN; the dynamic anchor box module generates adaptive anchor boxes in real time based on feature clustering or gradient guidance, ultimately outputting the OBB prediction.

[0039] In this application, LiteHead is integrated into the detection head of YOLOv11-OBB. By replacing the standard shared convolution and static anchor boxes, the feature processing pipeline is optimized, floating-point operations and memory accesses are reduced, and the ability to capture rotational details and global information is enhanced. The LiteHead architecture is well-suited for detecting and locating white leaf blight spots in rice paddy images captured by drones.

[0040] Despite the introduction of lightweight modules, the YOLOv11-OBB model still faces issues of parameter redundancy and computational intensity, leading to increased floating-point operations (GFLOPs) and memory requirements. This problem is particularly severe for resource-constrained UAV embedded systems, potentially causing inference latency and deployment bottlenecks.

[0041] To address these challenges, this application introduces a Layer Adaptive Amplitude Pruning (LAMP) strategy to further optimize the YOLOv11-OBB model. LAMP reduces the model's parameter size and computational cost through global pruning, making it more suitable for resource-constrained UAV platforms. LAMP is an amplitude-based pruning strategy that determines the pruning threshold by calculating an adaptive importance score (LAMP score) for each layer. It avoids the performance degradation caused by uniform sparsity and dynamically adjusts the sparsity of each layer based on the statistical characteristics of the weight distribution of each layer. Earlier layers retain more connections to capture fine-grained features, while later layers undergo more aggressive pruning to reduce computational complexity.

[0042] The LAMP pruning process includes the following steps: pre-training the complete model to obtain initial weights; calculating the global LAMP score, sorting all weights, and setting a threshold based on the target sparsity; applying amplitude-based pruning at each level to remove connections below the threshold; and fine-tuning the pruned model to restore accuracy, typically using techniques such as knowledge distillation or learning rate scheduling.

[0043] Compared to traditional amplitude pruning methods, LAMP maintains better accuracy after pruning by using a layer-adaptive mechanism. This allows for a significant reduction in parameters and GFLOPs without sacrificing detection performance. Figure 5 The LAMP process is demonstrated.

[0044] In this application, LAMP was applied to the backbone, neck, and detection head of YOLOv11-OBB, and after iterative pruning and fine-tuning, a lightweight variant was obtained. This optimized structure enhances resource utilization, supports the capture of detailed and global information from the perspective of UAVs, and is suitable for the detection and location of bacterial blight in rice. Example 2

[0045] Because drones move continuously while capturing video, some frames may not fully capture rice plants, and the same white leaf blight target may appear in multiple frames. Disease detection relying solely on a single image can lead to missed detections, false detections, or duplicate detections, thus affecting the accuracy and stability of the screening process. To address this issue, this application proposes a screening method. This method includes image acquisition and transmission, disease target detection, Kalman filter prediction, detection and prediction matching, disease counting, disease severity classification, and germplasm screening.

[0046] In image acquisition and transmission, the user controls the drone to fly, adjusts the camera's focal length and gimbal angle during flight, and then activates the recording function; the captured video data is transmitted to the edge computing device via wireless or wired connection.

[0047] In the detection of disease targets, the video recorded by the UAV is processed by the white leaf blight detection model proposed in this application to detect and generate the corresponding bounding boxes.

[0048] In Kalman filter prediction, a multi-target tracking model is introduced, and the Kalman filter is used to predict the target's position. The Kalman filter is a recursive estimation algorithm that uses statistical inference from historical data to estimate the target's current position. This prediction provides prior information for disease detection in subsequent frames, thereby reducing bias from occlusion or detection errors.

[0049] The Kalman filter state update equation is ,in, Let F represent the predicted state at time t, B be the state transition matrix, and u be the control input matrix. t This is the control vector. This prediction helps improve the accuracy of subsequent frame detection.

[0050] In the detection and prediction matching process, the Hungarian algorithm is used to match the detection results of the current frame with the positions predicted by the Kalman filter. The matching score is calculated using the following formula: ,in, The intersection-union ratio (IUR) of the bounding boxes. This represents the similarity of re-identified features, where λ is a weighting parameter that balances spatial overlap and appearance features.

[0051] In disease counting, white leaf blight targets are tracked in the list, and each target has a unique identifier used to count the number of times white leaf blight occurs.

[0052] In the disease severity classification, the disease incidence rate is calculated using a formula based on the disease count and the number of times rice has been inoculated. Then, based on the incidence rate, rice was divided into bacterial blight resistance levels, as shown in Table 1.

[0053] Table 1 Classification criteria for rice germplasm resistant to Bacterial Leaf Blight

[0054] In germplasm screening, L1 or L2 grade rice was selected as the preferred candidate material for bacterial blight resistance.

[0055] The above specific embodiments are merely several preferred embodiments of the present invention. Based on the technical solutions of the present invention and the relevant teachings of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. A method for screening resistance to bacterial blight in rice based on unmanned aerial vehicles and deep learning, characterized in that, The operating steps include: S1. Drone image acquisition: The drone flies along a predetermined flight route and at a predetermined altitude to capture images of white leaf blight spots in the rice paddies. S2. Detection and localization of bacterial blight spots: After inputting image data, the YOLOv11-OBB model is used to detect bacterial blight spots. The improved YOLOv11-OBB model consists of the C3k2FC module, SPPF_LSKA module, SlimNeck module, and LiteHead module, which are responsible for extracting features from the image and localizing bacterial blight spots. S3. Target Tracking and Multi-Target Recognition: Using a Kalman filter, multi-target tracking is performed on white leaf blight spots in each frame of the image to avoid occlusion and the repeated appearance of the same target in the image during flight. S4. Disease incidence rate calculation and disease resistance classification: The disease incidence rate of rice is calculated based on the number of detected lesions. The rice is then classified into different disease resistance levels using the threshold of the incidence rate, so as to screen rice varieties based on disease resistance. S5. Screening Results Output and Display: The final screening results are output and presented to users through a graphical interface or report to help breeding researchers make selections for disease-resistant varieties. In S2, the overall structure of the improved YOLOv11-OBB model includes a backbone, neck, and head. Improvements include: in the backbone network, the C3k2FC module integrates PConv and CGLU into the C3k2 structure; in the backbone network, the SPPF_LSKA module incorporates LSKA pooling; in the neck, the SlimNeck module uses GSConv and VoVGSCSP to optimize feature fusion while minimizing FLOPs; in the head, the LiteHead module employs separate batch normalization and dynamic anchor boxes; additionally, LAMP pruning implements adaptive global pruning to compress model parameters.

2. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: For the C3k2FC module, the two 1×1 point convolutions in the FasterNet block were replaced with the CGLU module, forming an improved FC structure; at the same time, PConv was integrated into the depthwise convolution segment of the bottleneck part of C3k2, ultimately forming the improved C3k2FC module.

3. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: For the SPPF_LSKA module, the structure first extracts multi-resolution features through multiple max pooling branches, and then applies the LSKA module to perform attention weighting on the concatenated features; the horizontal and vertical branches of LSKA process spatial information, and depthwise separable convolution further optimizes channel mixing; finally, the fused features are output through 1×1 convolution.

4. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: For the SlimNeck module, the GSConv unit introduces a Shuffle mechanism to fuse the channel-correlated data obtained from the SC with the spatial output from the DSC.

5. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: For the SlimNeck module, the VoVGSCSP unit divides the input matrix channels into two parts: one part undergoes convolutional preprocessing and features are extracted through a series of GS bottleneck units, while the other part serves as the residual path and undergoes only a single convolutional transformation.

6. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: The core components of the LiteHead detector include a shared convolutional layer, a separate BN module, and a dynamic anchor box generator.

7. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 1, characterized in that: LAMP determines the pruning threshold by calculating an adaptive importance score for each layer.

8. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 4, characterized in that: The core components of the GSConv unit include Conv, DWConv, Concat, and Shuffle operations.

9. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 5, characterized in that: The VoVGSCSP unit leverages the CSP framework to enhance feature reuse, minimizes data loss through a one-time aggregation method, and integrates batch normalization and the SiLU activation function to maintain feature consistency.

10. The method for screening rice bacterial blight resistance based on UAVs and deep learning according to claim 6, characterized in that: Shared convolutional layers are used for multi-task reuse; separate BN modules apply BN independently to each detection branch; and a dynamic anchor box generator calculates anchor box parameters in real time based on feature map statistics.

Images