Bursaphelenchus xylophilus disease detection method, system, medium and equipment

Abstract

The invention discloses a pine wood nematode disease detection method, system, medium and equipment, and belongs to the technical field of artificial intelligence and forest disease monitoring, the pine wood nematode disease detection method comprises the following steps: according to different infection stages of pine wood nematode disease, labeling a lesion area of an RGB image of a pine wood nematode disease color-changing diseased wood, and establishing a training set; and training a pine wood nematode disease change detection model by adopting the training set, so that the pine wood nematode disease change detection model learns features corresponding to the labels in different infection stages until the pine wood nematode disease change detection model can correctly identify the disease spot position of the infected wood and the corresponding infection stage. According to the method, the positions of pine wood nematode disease spots and the infection stage of the diseased wood can be accurately recognized, and the problems that local features are prone to being interfered and the recognition effect is poor when a current detection method faces early-stage tiny disease spots and background and shielding under a complex mixed forest are solved.

Description

Technical Field

[0001] This invention relates to a method, system, medium, and equipment for detecting pine wilt disease, belonging to the field of artificial intelligence and forest disease monitoring technology. Background Technology

[0002] Pine wilt disease, also known as pine blight, is caused by the pine wilt nematode. It is characterized by rapid spread and high mortality, and has caused significant ecological and economic losses in many countries and regions of my country. Practice has shown that early detection and localization are key to curbing its spread and reducing losses.

[0003] Current monitoring methods primarily rely on patrols in planted forest areas, but limitations imposed by complex mountainous terrain, road conditions, and operational costs make it difficult to meet the demands for large-scale, near-real-time, and precise monitoring. In contrast, unmanned aerial vehicle (UAV) platforms offer advantages such as wide coverage, high efficiency, and low cost, enabling them to quickly acquire high-resolution aerial images and providing a data foundation for the automatic identification of forest pests and diseases. Meanwhile, the development of deep learning and computer vision has made it possible to identify early subtle symptoms from images. In terms of algorithms, convolutional neural network (CNN) methods perform well in general object detection, but they have inherent shortcomings in long-range dependencies and global semantic modeling. When faced with early subtle lesions, complex mixed forest backgrounds, and occlusions, local features are easily disturbed. Although single-stage detectors (such as the YOLO series) have good real-time performance, they usually rely on non-maximum suppression (NMS) and manual hyperparameter trade-offs, and their adaptability and ability to characterize small objects in complex scenes are limited.

[0004] DETR-like end-to-end detection paradigms model detection as set predictions, combining Transformer and Hungarian matching to significantly simplify candidate boxes and post-processing. However, its slow convergence, weak perception of small targets, insufficient representation of local details, and limited inference efficiency restrict its application in forestry scenarios with limited resources and high real-time requirements. Subsequent improvements such as RT-DETR improve fusion and initialization quality and shorten training and inference paths through hybrid encoders and IoU perception, but still have shortcomings in the stable characterization of complex boundaries, shape representation, and small-scale lesions. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, system, medium and equipment for detecting pine wilt disease, which can accurately identify the location of pine wilt disease spots and the infection stage of infected trees. It solves the problem that the current detection method is easily interfered with by local features and has poor identification effect when facing early fine lesions, complex mixed forest background and shading.

[0006] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution: The first aspect of this invention provides a method for detecting pine wilt disease, comprising: Obtain the RGB image of the target pine forest; The RGB images of the target pine forest are input into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of infected trees in the RGB images of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

[0007] Furthermore, the pine wilt disease detection model includes: The feature extraction module includes multiple convolutional layers, which are used to extract enhanced features from the RGB image multiple times, taking into account both spatial resolution and semantic level. The first fusion module is used to upsample and fuse the enhanced features extracted multiple times to obtain multi-level first fusion features; The second fusion module is used to perform downsampling fusion based on the multi-level first enhanced features to obtain the multi-level second fusion features; The detection module is used to obtain the recognition result based on the first fusion feature and the second fusion feature.

[0008] Furthermore, the feature extraction module includes a channel projection layer and multiple layers of improved convolutional modules connected in sequence; the channel projection layer is connected to the first layer of the improved convolutional module. The improved convolution module includes a wavelet domain convolution branch connected to the channel projection layer for enhancing local details and texture features in the frequency domain, and a spatial domain convolution branch for enhancing spatial texture and edge response.

[0009] Furthermore, the improved convolutional module is represented as: ; ; ; in, Represents an RGB image. This indicates channel-aligned convolution processing with a size of [size missing]. , Indicates the output of the channel projection layer; This represents the initial output features of the first improved convolutional module; This indicates attention fusion processing. Representing augmented detail convolution operations, This indicates inverse wavelet transform processing. This represents the bottom-up wavelet domain inverse reconstruction operation. This indicates the number of wavelet domain convolutional branches. This indicates channel-scale modulation operation. This indicates a depthwise separable convolution operation. This indicates wavelet transform processing; Indicates the first Enhanced features output by layer-by-layer improved convolutional modules No. Enhanced features are added to the output of the improved convolutional module. This indicates attention aggregation processing. This indicates channel splicing processing. Indicates the first Improved convolution module processing.

[0010] Furthermore, the first fusion module includes a first upsampling fusion module and a multi-layer second upsampling fusion module; the first upsampling fusion module is connected to the first-layer second upsampling fusion module; The first upsampling fusion module is represented as: ; This represents the first fusion feature output by the first upsampling fusion module; This indicates a deserialization operation; This indicates normalization processing. Indicates the first Enhanced features are added to the output of the improved convolutional module. This indicates convolution processing, with a size of [size missing]. The step size is 1; This indicates convolution processing, with a size of [size missing]. The step size is 1; This represents a learnable location code; This indicates regularization processing. This indicates multi-head attention processing. This indicates feedforward convolution processing; The second upsampling fusion module is represented as follows: ; ; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; This refers to the RepNCSPELAN module, used for feature enhancement and remodeling; This indicates the 1×1 channel compression and alignment after merging; This indicates the CGAFusion fusion module; Indicates the upsampling operator; This represents a horizontal 1×1 convolution; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer.

[0011] Furthermore, the second fusion module includes multiple layers of downsampling fusion modules connected in sequence; The downsampling fusion module is represented as follows: ; ; This represents the second fusion feature output by the first-layer downsampling fusion module; This indicates the RepNCSPELAN module, used for feature enhancement and remodeling; This represents a horizontal 1×1 convolution; This indicates the CGAFusion fusion module; This indicates convolution processing, with a size of [size missing]. The step size is 1; This indicates downsampling processing, with a size of [size missing]. The number of samplings is Second-rate; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; Indicates the first The second fusion feature output by the second upsampling fusion module of the second layer; This represents the second fused feature output by the second-layer downsampling fusion module. This represents the second fusion feature output by the first upsampling fusion module.

[0012] Furthermore, a training set was used to train the pine wilt disease detection model, enabling it to learn the features corresponding to labels of different infection stages, until the model could correctly identify the location of lesions and the corresponding infection stage in infected trees, including: The training process is optimized using classification loss and bounding box regression loss, as follows: ; in, Represents the total loss function; Represents the classification loss function. Represents the classification loss coefficient; This represents the bounding box regression loss function. Indicates the regression loss coefficient; ; This indicates the probability that the model's output belongs to the foreground target. This represents the intersection-union ratio (IoU) between the predicted bounding box and its corresponding ground truth bounding box. Labels indicating different stages of infection. This indicates that the predicted result matched the actual result successfully. This indicates the image background or a non-matching sample. Indicates the focus factor; ; Indicates confidence loss. Indicates the confidence loss coefficient; Indicates the first The location information of the actual bounding box. , Indicates the first The x-coordinate of the center point of the true bounding box Indicates the first The y-coordinate of the center point of the true bounding box Indicates the first The width of a real bounding box. Indicates the first The height of the actual bounding box; This represents the L1 coordinate regression loss. Indicates the L1 coordinate regression coefficients; Indicates the relationship with the first The location information of the predicted bounding boxes obtained by Hungarian matching of the true bounding boxes. Indicates the first The prediction index corresponding to each true bounding box.

[0013] A second aspect of the present invention provides a pine wilt disease detection system, comprising: The acquisition module is used to acquire RGB images of the target pine forest. The detection module is used to input the RGB image of the target pine forest into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of the infected trees in the RGB image of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

[0014] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the pine wilt disease detection method as described above.

[0015] A fourth aspect of the present invention also provides a computer device, comprising: Memory, used to store instructions; A processor is configured to execute the instructions, causing the device to perform the pine wilt disease detection method as described above.

[0016] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: 1. This invention labels the lesion areas of RGB images of pine wilt disease-infected trees according to different infection stages, and establishes a training set. The training set is used to train a pine wilt disease detection model, enabling the model to learn the features corresponding to the labels of different infection stages, until the model can correctly identify the location of lesions and the corresponding infection stage of the infected trees. This solves the problem that current detection methods are prone to interference with local features and have poor recognition results when faced with early, subtle lesions, complex mixed forest backgrounds, and occlusion.

[0017] 2. The improved convolution module of the present invention includes a wavelet domain convolution branch connected to the channel projection layer for enhancing local details and texture features in the frequency domain and a spatial domain convolution branch for enhancing spatial texture and edge response. The improved convolution module is used to process RGB images and extract enhanced features that take into account both spatial resolution and semantic level, thereby improving the accuracy of identifying various infection stages of infected trees, especially early subtle lesions. Attached Figure Description

[0018] Figure 1 This is a flowchart of the training process for the pine wood nematode lesion detection model provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the pine wilt disease detection model provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the improved convolution module in the pine wilt disease detection model provided in this embodiment of the invention. Detailed Implementation

[0019] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention. Example 1

[0020] A method for detecting pine wilt disease includes: Obtain the RGB image of the target pine forest; The RGB images of the target pine forest are input into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of infected trees in the RGB images of the target pine forest. like Figure 1 As shown, the training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas in the RGB images of pine wilt diseased wood were labeled to establish a training set; specifically: Using drones equipped with high-resolution visible light cameras, multi-temporal aerial photography was conducted on the target forest area to collect images of pine forests under different lighting, seasonal, and perspective conditions. After the acquired RGB images were processed by slicing, denoising and brightness equalization, a standard pine forest RGB image dataset was obtained. The standard pine forest RGB image dataset was labeled using the LabelImg tool. Rectangular boxes were used to mark the lesion areas in the images according to the different infection stages of early, middle, late and necrotic stages, forming labels for the corresponding early, middle, late and necrotic stages. The labels were stored in ".XML" file format to obtain the training set.

[0021] The pine wilt disease detection model was trained using a training set, enabling the model to learn the features corresponding to the labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees. like Figure 2 As shown, the pine wilt disease detection model includes: The feature extraction module includes multiple convolutional layers, used to extract enhanced features from the RGB image multiple times, taking into account both spatial resolution and semantic level; the feature extraction module includes a channel projection layer and multiple sequentially connected improved convolutional modules; the channel projection layer is connected to the first improved convolutional module. like Figure 3 As shown, the improved convolutional module includes wavelet domain convolution branches connected to the channel projection layers for enhancing local details and texture features in the frequency domain, and spatial domain convolution branches for enhancing spatial texture and edge response; specifically, the improved convolutional module is represented as follows: ; ; ; in, Represents an RGB image. This indicates channel-aligned convolution processing with a size of [size missing]. , Indicates the output of the channel projection layer; This represents the initial output features of the first improved convolutional module; This indicates attention fusion processing. Representing augmented detail convolution operations, This indicates inverse wavelet transform processing. This represents the bottom-up wavelet domain inverse reconstruction operation. This indicates the number of wavelet domain convolutional branches. Indicates from the first From the first floor to the second floor; Indicates channel scale modulation operation; This represents a depthwise separable convolution operation, corresponding to the one in the diagram. ; This represents wavelet transform processing; the figure shows... Indicates feature fusion operation; Indicates the first Enhanced features output by layer-by-layer improved convolutional modules No. Enhanced features are added to the output of the improved convolutional module. This indicates attention aggregation processing. This indicates channel splicing processing. Indicates the first Improved convolution module processing.

[0022] The first fusion module is used to upsample and fuse the enhanced features extracted multiple times to obtain multi-level first fusion features. The first fusion module includes a first upsampling fusion module and a multi-level second upsampling fusion module. The first upsampling fusion module is connected to the first-level second upsampling fusion module. Specifically, the first upsampling fusion module is represented as follows:

[0023] ; This represents the first fusion feature output by the first upsampling fusion module; This indicates a deserialization operation; This indicates normalization processing; This indicates convolution processing, with a size of [size missing]. The step size is 1; This indicates convolution processing, with a size of [size missing]. The step size is 1; This represents a learnable location code; This indicates regularization processing. This indicates multi-head attention processing. This indicates feedforward convolution processing; The second upsampling fusion module is represented as follows: ; ; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; This refers to the RepNCSPELAN module, used for feature enhancement and remodeling; This indicates the 1×1 channel compression and alignment after merging; This indicates the CGAFusion fusion module; Indicates the upsampling operator; This represents a horizontal 1×1 convolution used for channel projection and scale alignment of features from deeper layers; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer.

[0024] The second fusion module is used to perform downsampling fusion based on the multi-level first enhanced features to obtain multi-level second fused features; the second fusion module includes multiple layers of downsampling fusion modules connected in sequence; specifically, the downsampling fusion module is represented as follows: ; ; This represents the second fusion feature output by the first-layer downsampling fusion module; This indicates the RepNCSPELAN module, used for feature enhancement and remodeling; This indicates the CGAFusion fusion module; This indicates downsampling processing, with a size of [size missing]. The number of samplings is Second-rate; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; Indicates the first The second fusion feature output by the second upsampling fusion module of the second layer; This represents the second fused feature output by the second-layer downsampling fusion module. This represents the second fused feature output by the first upsampling fusion module; The detection module is used to obtain the recognition result based on the first fusion feature and the second fusion feature.

[0025] The training process is optimized using classification loss and bounding box regression loss, as follows: ; in, Represents the total loss function; Represents the classification loss function. Represents the classification loss coefficient; This represents the bounding box regression loss function. Indicates the regression loss coefficient; ; This indicates the probability that the model's output belongs to the foreground target. This represents the intersection-union ratio (IoU) between the predicted bounding box and its corresponding ground truth bounding box. Labels indicating different stages of infection. This indicates that the predicted result matched the actual result successfully. This indicates the image background or a non-matching sample. Indicates the focus factor; ; Indicates confidence loss. Indicates the confidence loss coefficient; Indicates the first The location information of the actual bounding box. , Indicates the first The x-coordinate of the center point of the true bounding box Indicates the first The y-coordinate of the center point of the true bounding box Indicates the first The width of a real bounding box. Indicates the first The height of the actual bounding box; This represents the L1 coordinate regression loss. Indicates the L1 coordinate regression coefficients; Indicates the relationship with the first The location information of the predicted bounding boxes obtained by Hungarian matching of the true bounding boxes. Indicates the first The prediction index corresponding to each true bounding box.

[0026] To verify the superiority of the method proposed in this embodiment, a comparative experiment was conducted using the following existing target detection models: Single Shot MultiBox Detector (SSD); A classic two-stage object detection model (Faster Region-Based Convolutional Neural Network, Faster R-CNN). Two lightweight one-stage detection models (You Only Look Once version 8 nano / version 12 nano, YOLOv8n and YOLOv12n). A lightweight convolutional neural network, MobleNetv3, designed for mobile devices; A lightweight convolutional neural network based on channel shuffling: ShuffleNetv2; A real-time one-stage target detection model based on Transformer (Real-Time DetectionTransformer version 2 with ResNet-18 backbone, RT-DETRv2-R18). A lightweight real-time DETR target detection model based on improved matching paradigm and fine-grained distribution regression (Detection with Improved Matching combined with Fine-grained Distribution Refinement in DETRs (Nano version), DEIM-D-Fine-N).

[0027] The test metrics include average accuracy (AP), frames per second (FPS), parameters, and floating point operations per second (FLOPs). in, ; Represents the probability density function; Indicates precision. ; Indicates recall rate, ; In this context, TP (True Positive) represents a correct positive detection, which is the bounding box of pine wood lesions correctly predicted by the model; FP (False Positive) represents a false positive detection, which is the model incorrectly detecting background or healthy pine wood as diseased wood; and FN (False Negative) represents a missed positive target, which is the diseased wood that actually exists in the image but was not detected by the model. ; This represents the time it takes for the model to process one image. ; This indicates the number of output channels of the model. This indicates the number of channels in the input feature map. Indicates the height of the convolution kernel. Indicates the width of the convolution kernel; ; Indicates the height of the input feature map. Input the width of the feature map; The experimental results are shown in Table 1:

[0028] It should be noted that in Table 1, AP50 represents the average precision value calculated when the IoU threshold is 0.50; PWD-E represents the early stage label of pine wilt disease infection, PWD-M represents the mid-stage label of pine wilt disease infection, PWD-L represents the late stage label of pine wilt disease infection, and PWD-D represents the wilt stage label of pine wilt disease infection; All represents the combined average AP50 of PWD-E, PWD-M, PWD-L, and PWD-D. As shown in Table 1, the method proposed in this embodiment not only achieves an accuracy of 83.8% in early-stage disease detection and an overall detection accuracy of 90.2%, but also has fewer model parameters and higher real-time performance, achieving a balance between speed and accuracy. Its performance in all aspects is superior to existing target detection models.

[0029] Example 2 A pine wilt disease detection system, comprising: The acquisition module is used to acquire RGB images of the target pine forest. The detection module is used to input the RGB image of the target pine forest into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of the infected trees in the RGB image of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

[0030] Example 3 A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following method steps: Obtain the RGB image of the target pine forest; The RGB images of the target pine forest are input into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of infected trees in the RGB images of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

[0031] Example 4 A computer device, comprising: Memory, used to store instructions; A processor, used to execute instructions that cause a device to perform the following method steps: Obtain the RGB image of the target pine forest; The RGB images of the target pine forest are input into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of infected trees in the RGB images of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

[0032] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0033] This application is described with reference to flowchart illustrations of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each step in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 One or more processes or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0034] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 The function specified in one or more processes.

[0035] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 Steps of a specified function in one or more processes.

[0036] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for detecting pine wilt disease, characterized in that, include: Obtain the RGB image of the target pine forest; The RGB images of the target pine forest are input into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of infected trees in the RGB images of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

2. The method for detecting pine wilt disease according to claim 1, characterized in that, The pine wilt disease detection model includes: The feature extraction module includes multiple convolutional layers, which are used to extract enhanced features from the RGB image multiple times, taking into account both spatial resolution and semantic level. The first fusion module is used to upsample and fuse the enhanced features extracted multiple times to obtain multi-level first fusion features; The second fusion module is used to perform downsampling fusion based on the multi-level first enhanced features to obtain the multi-level second fusion features; The detection module is used to obtain the recognition result based on the first fusion feature and the second fusion feature.

3. The method for detecting pine wilt disease according to claim 2, characterized in that, The feature extraction module includes a channel projection layer and multiple layers of improved convolutional modules connected in sequence; the channel projection layer is connected to the first layer of the improved convolutional module. The improved convolution module includes a wavelet domain convolution branch connected to the channel projection layer for enhancing local details and texture features in the frequency domain, and a spatial domain convolution branch for enhancing spatial texture and edge response.

4. The method for detecting pine wilt disease according to claim 3, characterized in that, The improved convolutional module is represented as follows: ； ； ； in, Represents an RGB image. This indicates channel-aligned convolution processing with a size of [size missing]. , Indicates the output of the channel projection layer; This represents the initial output features of the first improved convolutional module; This indicates attention fusion processing. Representing augmented detail convolution operations, This indicates inverse wavelet transform processing. This represents the bottom-up wavelet domain inverse reconstruction operation. This indicates the number of wavelet domain convolutional branches. This indicates channel-scale modulation operation. This indicates a depthwise separable convolution operation. This indicates wavelet transform processing; Indicates the first Enhanced features output by layer-by-layer improved convolutional modules No. Enhanced features are added to the output of the improved convolutional module. This indicates attention aggregation processing. This indicates channel splicing processing. Indicates the first Improved convolution module processing.

5. The method for detecting pine wilt disease according to claim 2, characterized in that, The first fusion module includes a first upsampling fusion module and a multi-layer second upsampling fusion module; the first upsampling fusion module is connected to the first-layer second upsampling fusion module. The first upsampling fusion module is represented as: ； This represents the first fusion feature output by the first upsampling fusion module; This indicates a deserialization operation; This indicates normalization processing. Indicates the first Enhanced features are added to the output of the improved convolutional module. This indicates convolution processing, with a size of [size missing]. The step size is 1; This indicates convolution processing, with a size of [size missing]. The step size is 1; This represents a learnable location code; This indicates regularization processing. This indicates multi-head attention processing. This indicates feedforward convolution processing; The second upsampling fusion module is represented as follows: ； ； Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; This refers to the RepNCSPELAN module, used for feature enhancement and remodeling; This indicates the 1×1 channel compression and alignment after merging; This indicates the CGAFusion fusion module; Indicates the upsampling operator; This represents a horizontal 1×1 convolution; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer.

6. The method for detecting pine wilt disease according to claim 2, characterized in that, The second fusion module includes multiple layers of downsampling fusion modules connected in sequence; The downsampling fusion module is represented as follows: ； ； This represents the second fusion feature output by the first-layer downsampling fusion module; This refers to the RepNCSPELAN module, used for feature enhancement and remodeling; This represents a horizontal 1×1 convolution; This indicates the CGAFusion fusion module; This indicates convolution processing, with a size of [size missing]. The step size is 1; This indicates downsampling processing, with a size of [size missing]. The number of samplings is Second-rate; Indicates the first The first fusion feature output by the second upsampling fusion module of the layer; Indicates the first The second fusion feature output by the second upsampling fusion module of the second layer; This represents the second fused feature output by the second-layer downsampling fusion module. This represents the second fusion feature output by the first upsampling fusion module.

7. The method for detecting pine wilt disease according to claim 1, characterized in that, A pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages, until the model could correctly identify the location of lesions and the corresponding infection stage in infected trees, including: The training process is optimized using classification loss and bounding box regression loss, as follows: ； in, Represents the total loss function; Represents the classification loss function. Represents the classification loss coefficient; This represents the bounding box regression loss function. Indicates the regression loss coefficient; ； This indicates the probability that the model's output belongs to the foreground target. This represents the intersection-union ratio (IoU) between the predicted bounding box and its corresponding ground truth bounding box. Labels indicating different stages of infection. This indicates that the predicted result matched the actual result successfully. This indicates the image background or a non-matching sample. Indicates the focus factor; ； Indicates confidence loss. Indicates the confidence loss coefficient; Indicates the first The location information of the actual bounding box. , Indicates the first The x-coordinate of the center point of the true bounding box Indicates the first The y-coordinate of the center point of the true bounding box Indicates the first The width of a real bounding box. Indicates the first The height of the actual bounding box; This represents the L1 coordinate regression loss. Indicates the L1 coordinate regression coefficients; Indicates the relationship with the first The location information of the predicted bounding boxes obtained by Hungarian matching of the true bounding boxes. Indicates the first The prediction index corresponding to each true bounding box.

8. A pine wilt disease detection system, characterized in that, include: The acquisition module is used to acquire RGB images of the target pine forest. The detection module is used to input the RGB image of the target pine forest into a pre-trained pine wilt disease detection model to determine the location of lesions and the infection stage of the infected trees in the RGB image of the target pine forest. The training of the pine wilt disease detection model includes: According to the different infection stages of pine wilt disease, the lesion areas of RGB images of pine wilt diseased wood were labeled to establish a training set; The pine wilt disease detection model was trained using a training set, enabling it to learn the features corresponding to labels of different infection stages until it could correctly identify the location of lesions and the corresponding infection stage in infected trees.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the pine wilt disease detection method as described in any one of claims 1 to 7.

10. A computer device, characterized in that, include: Memory, used to store instructions; A processor for executing the instructions, causing the device to perform the pine wilt disease detection method as described in any one of claims 1 to 7.

Images