Bursaphelenchus xylophilus disease identification method and device based on inspection view angle of unmanned aerial vehicle, and medium

Abstract

The invention provides a pine wood nematode disease identification method and device based on the inspection view angle of an unmanned aerial vehicle and a medium, and belongs to the field of forestry disease monitoring and prevention, and the method comprises the steps: setting the height, flight speed and shooting angle of the unmanned aerial vehicle, and obtaining an image shot by the unmanned aerial vehicle in the inspection process in a preset region; inputting the image into an insect disease detection model to obtain an insect disease detection result; wherein the insect disease detection model is obtained by training in a mode of adjusting model parameters by adopting a stochastic gradient descent optimizer and a complete intersection-to-union ratio loss function. According to the identification method provided by the embodiment of the invention, data preprocessing is performed on the picture shot by the unmanned aerial vehicle, the pine wood nematode disease is automatically identified by using the deep learning algorithm, and manual field detection is replaced by automatic identification, so that the problems that manual inspection consumes a large amount of manpower and is influenced by the environment are avoided, the manpower resource waste is reduced, and the detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of forestry disease monitoring and prevention, and in particular to a pine wood nematode disease identification method, equipment and medium based on the inspection perspective of an unmanned aerial vehicle. Background Art

[0002] With the advancement of modern agriculture and forestry technology, the protection and management of forest resources are developing in the direction of intelligence and precision. Pest and disease problems are important issues of global concern, especially pine wood nematode disease, which is a major forestry disease that has attracted much attention. Once pine trees are infected, the disease will spread rapidly, causing tree death, threatening the stability of the ecosystem and the safety of forest resources. At present, the prevention and control of pine wood nematode disease mainly relies on manual inspections, laboratory analysis, and disease diagnosis methods of professional technicians. These methods are inefficient, prone to missed detections and false detections, and time-consuming, and cannot meet the needs of large-scale, real-time monitoring.

[0003] In recent years, drone technology has been increasingly used in agriculture and forestry, becoming an important tool for remote sensing monitoring, vegetation analysis, and pest control. Drones have the characteristics of long range, low cost, and simple operation, and can efficiently cover large areas of forest for monitoring. However, how to achieve automated analysis and accurate recognition of drone-collected images remains a major challenge in this field.

[0004] At present, the traditional manual inspection method requires a lot of manpower and time, it is difficult to cover large forest areas, and it is easily affected by factors such as weather and terrain, resulting in incomplete inspections and a high risk of missed detection and false detection. Existing disease diagnosis methods usually rely on laboratory analysis, which is time-consuming and cannot meet the needs of rapid response and timely prevention and control. In addition, existing monitoring methods are difficult to achieve real-time and continuous disease monitoring. The disease is not discovered until it spreads, missing the best time for prevention and control. Existing recognition methods have poor adaptability to different environmental conditions and disease manifestations, and the recognition accuracy is unstable. Therefore, by combining drone inspections and deep learning target detection technology, it is necessary to establish an efficient, accurate, and real-time automatic identification method for pine wood nematode disease to improve the overall level of disease monitoring and prevention and control.

[0005] These problems and defects are known. In the prevention and control of pine wood nematode disease in forestry, traditional manual inspections and diagnostic methods that rely on laboratory analysis have long been used, which require a lot of manpower and time, are easily affected by weather and terrain factors, and have serious missed detections and false detections. Existing disease diagnosis methods are too time-consuming and difficult to respond quickly. At the same time, existing monitoring methods are unable to achieve real-time and continuous monitoring, resulting in the problem of missing the best time for prevention and control. In addition, the poor adaptability and unstable accuracy of existing identification methods under different environmental conditions and disease manifestations have also been fully exposed in practice. These are existing problems that are well known in long-term practice. Summary of the invention

[0006] In view of this, the present invention provides a pine wood nematode disease identification method, equipment and medium based on the drone inspection perspective. The method can quickly detect the occurrence and development of pine wood nematode disease in a large area, identify disease symptoms in time, and provide a basis for scientific decision-making.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] According to an embodiment of the present invention, a method for identifying pine wood nematode disease based on the inspection perspective of a drone includes: setting the height, flight speed and shooting angle of the drone, and obtaining images taken by the drone during the inspection process in a preset area; inputting the images into a pest detection model to obtain pest detection results; wherein the pest detection model is trained by adjusting model parameters using a stochastic gradient descent optimizer and a complete intersection-over-union loss function.

[0009] As an embodiment of the first aspect of the present invention, the training method of the insect disease detection model includes:

[0010] Construct the original image obtained from the drone inspection perspective;

[0011] Preprocessing the original image to obtain an insect disease detection data set;

[0012] The target detection model is trained based on the insect disease detection data. During the training process, the parameters of the target detection model are adjusted using a stochastic gradient descent optimizer and a complete intersection-over-union loss function, and the target detection model is trained to obtain the insect disease detection model.

[0013] As an embodiment of the first aspect of the present invention, the preprocessing of the original image includes:

[0014] Cutting the original image of the original size into a plurality of image slices, removing similar and identical images;

[0015] Each cut image is manually labeled to mark the location and category of pine wood nematode disease, thereby obtaining the pine disease detection data set.

[0016] As an embodiment of the first aspect of the present invention, the preprocessing of the original image further includes:

[0017] The annotated original image is enlarged, wherein the enlargement method includes at least one of cutting enhancement, random flipping, center cropping, random brightness transformation, random contrast transformation, random saturation transformation, adding Gaussian noise and mixed enhancement.

[0018] As an embodiment of the first aspect of the present invention, the target detection model includes:

[0019] A backbone network, wherein the backbone network includes convolution, deconvolution, residual connection and bottleneck structure, and the backbone network uses a spatial pyramid pooling module to extract multi-scale features in the image;

[0020] A neck network, used for fusing the extracted features at multiple scales;

[0021] The head network,adopts the design of decoupled head to perform prediction,classification and regression on the features.

[0022] As an embodiment of the first aspect of the present invention, the neck network adopts a feature pyramid network and a path aggregation network structure to fuse multiple scale feature maps.

[0023] As an embodiment of the first aspect of the present invention, the training method of the insect disease detection model further includes: screening out the final detection result by a non-maximum suppression method, removing overlapping detection frames, and retaining the most likely detection result.

[0024] As an embodiment of the first aspect of the present invention, the training method of the insect disease detection model further includes:

[0025] During the training of the target detection model, the trained model is evaluated, and the evaluation uses precision, recall, and average precision as the performance evaluation criteria for the target detection algorithm.

[0026] In a second aspect, the present invention further discloses an electronic device, comprising: a memory for storing instructions executed by one or more processors of the electronic device, and a processor for executing the identification method described in the embodiment of the first aspect.

[0027] In a third aspect, the present invention further discloses a computer-readable storage medium, on which instructions are stored. When the instructions are executed on an electronic device, the electronic device executes the identification method described in the embodiment of the first aspect.

[0028] The above technical solution of the present invention has at least one of the following beneficial effects:

[0029] According to the pine wood nematode disease identification method based on the perspective of drone inspection in an embodiment of the present invention, a stochastic gradient descent optimizer and a complete intersection-over-union loss function are used to train the model, so that the model can learn the position and shape information of the target more comprehensively during the training process. In the detection of pine wood nematode disease, the positioning of the boundary of the infected area will be more accurate, and infected and uninfected trees can be better distinguished, thereby effectively improving the accuracy of target positioning. In addition, by using the pictures obtained by drone inspection, by adjusting the flight parameters during the flight of the drone, images that are more conducive to the discovery of insect diseases can be obtained, which is conducive to improving the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a flow chart of a method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to an embodiment of the present invention;

[0031] Figure 2 A flowchart of a method for training an insect disease detection model according to an embodiment of the present invention;

[0032] Figure 3 Flow chart of another method for training an insect disease detection model according to an embodiment of the present invention

[0033] Figure 4 A structural diagram of fast spatial pyramid pooling according to an embodiment of the present invention;

[0034] Figure 5 Schematic diagram of a feature pyramid network and a path aggregation network structure according to an embodiment of the present invention.

[0035] Figure 6 FIG. 4 is a block diagram of a system on chip according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0037] To facilitate understanding of the solutions of the embodiments of the present application, the technical problems to be solved by the embodiments of the present application are first analyzed.

[0038] The following first describes in detail the pine wood nematode disease identification method based on the drone inspection perspective according to an embodiment of the present invention with reference to the accompanying drawings.

[0039] Figure 1The flowchart of the method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to an embodiment of the present invention is shown. Figure 1 As shown, a method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to an embodiment of the present invention may include:

[0040] S110, obtaining images captured by the drone during inspection in a preset area.

[0041] In an embodiment of the present invention, the preset area may be an area known in advance to be likely to have insect diseases. In an embodiment of the present invention, pine wood nematode disease is used as an example for illustration. The drone can be equipped with a high-resolution camera, for example, a high-resolution visible light camera / camera, a three-dimensional lidar scanning device, a multispectral imager, and one or more of a monocular camera. In the process of collecting images, a suitable flight altitude, flight speed, and shooting angle are set to ensure the quality and coverage of the collected images. This image has a higher resolution, is more conducive to clearly distinguishing whether there are pine wood nematodes, helps the model accurately identify pine wood nematodes in the captured images, and is conducive to early detection of pine wood nematodes.

[0042] S120, inputting the image into an insect disease detection model to obtain an insect disease detection result.

[0043] Among them, the insect disease detection model uses stochastic gradient descent optimizer (SGD) and complete intersection-over-union loss (CIoU Loss) to adjust the model parameters.

[0044] Among them, the SGD optimizer effectively improved the training efficiency and stability, and the CIoU loss function further enhanced the positioning accuracy of the model, thereby significantly improving the high-precision recognition performance, optimizing the model's detection ability for pine wood nematode disease, and solving the problem of insufficient model performance.

[0045] CIoU Loss mainly solves the shortcomings of the original IoU loss function in bounding box regression, especially when dealing with objects with mismatched shapes and large position deviations. On the one hand, the Euclidean distance between the center points of the boxes is introduced as part of the loss. This helps to optimize the center position of the target so that the regression does not only rely on the intersection of the boxes, but also considers the spatial layout and relative position of the boxes. By penalizing the distance between the center points of the bounding boxes, the center position of the predicted box is closer to the true box. On the other hand, an additional term is introduced to measure the shape consistency (such as aspect ratio) between the two bounding boxes. This allows it to better optimize the aspect ratio of the box during the regression process, rather than just the position and size, thereby avoiding the problem of aspect ratio mismatch.

[0046] The pine wood nematode disease identification method based on the drone inspection perspective of the embodiment of the present invention performs data preprocessing on pictures taken by the drone and uses a deep learning algorithm to automatically identify pine wood nematode disease. Automatic identification replaces manual on-site inspection, avoiding the problem of manual inspection consuming a lot of manpower and being affected by the environment, reducing waste of human resources, and improving detection efficiency.

[0047] It can be understood that the identification method of the embodiment of the present invention can be applied to electronic devices, including but not limited to mobile phones, smart TVs, wearable devices, tablet computers, desktop computers, computers, wearable devices (for example, smart watches, smart glasses, helmets), virtual reality (VR) devices, augmented reality (AR) devices, wireless devices in industrial control, wireless devices in self-driving, wireless devices in smart grids, wireless devices in transportation safety, wireless devices in smart cities, wireless devices in smart homes, servers, and the like.

[0048] The training process of the insect disease detection model according to the embodiment of the present invention is described in detail below in conjunction with specific embodiments.

[0049] refer to Figure 2 and Figure 3 , Figure 2 A flow chart of a method for training an insect disease detection model according to an embodiment of the present invention is shown. Figure 3 FIG. 2 is a flow chart showing another method for training an insect disease detection model according to an embodiment of the present invention. Figure 2 As shown, the training method includes:

[0050] S210, obtaining an original image obtained from the inspection perspective of the drone.

[0051] This step is the process of data set collection. UAVs equipped with high-resolution cameras are used to inspect areas where pine wood nematode disease may exist, and then raw images are collected. During the collection process, appropriate flight altitude, speed, and shooting angle are set to ensure the quality and coverage of the collected images. In some embodiments, some images can also be obtained through the network to increase the number of samples.

[0052] S220, pre-processing the original images collected by the drone to obtain an insect disease detection dataset. This step corresponds to Figure 3In S310 and S320, S310 obtains the pine tree image from the drone and performs image preprocessing, image slicing, image de-coincidence, image format conversion, etc. S320 enhances the image.

[0053] In an embodiment of the present invention, preprocessing may include cutting a large-size original image into multiple smaller image slices to remove similar or even identical images. Use the graphical user interface labelimg tool for image annotation to manually annotate each image after cutting, and the annotation process requires one-to-one correspondence, and each image accurately annotates the position and category of pine wood nematode disease. The annotated original image data is amplified, wherein one or more of 8 data amplification enhancement methods may be included, for example, mixed enhancement MixUp, cutting enhancement CutMix, random flipping, center cropping, random brightness conversion, random contrast conversion, random saturation conversion, and adding Gaussian noise. In some embodiments, taking a total of 5417 photos of the enhanced data set as an example, the pictures are divided into training set, verification set, and test set according to the ratio of 8:1:1, and the training set is 4333, the verification set is 541, and the test set is 542 after division. The richness of the pine wood nematode disease data set is improved, reliable data is provided for model training, it is helpful to improve the effectiveness of the model, and overcome the lack of precision caused by data problems in existing detection methods.

[0054] In some embodiments, in addition to the above eight data enhancement methods, data enhancement techniques such as random rotation, affine transformation, elastic deformation, etc. can also be added. These methods can also further increase the diversity of the data set, enable the model to learn more image change patterns, and improve the generalization ability of the model.

[0055] S230, training the target detection model based on the insect disease detection dataset. Figure 3 In S330, a deep object detection model is constructed, including steps such as image feature extraction, image feature fusion, and CIOU loss function training.

[0056] The prepared dataset is input into the YOLOv8 target detection model for training. During the training process, appropriate training parameters are set, the input image size is 640×640 resolution, the learning rate is 0.01, the batch size is 32, and the number of iterations is 800. When extracting features from the backbone layer of the backbone network, convolution and deconvolution layers are used, and residual connections and bottleneck structures are combined to reduce the size of the network and enhance the ability of feature extraction. Figure 4The structure diagram of the fast spatial pyramid pooling shown in FIG. 1 uses the fast spatial pyramid pooling (Spatial Pyramid Pooling Fast, SPPF) module to perform pooling operations of different scales on the input feature map to obtain richer and more representative feature information.

[0057] refer to Figure 5 , Figure 5 A schematic diagram showing a feature pyramid network and a path aggregation network structure according to an embodiment of the present invention is shown.

[0058] like Figure 5 As shown in the figure, when the neck layer of the network performs multi-scale feature fusion, the Feature Pyramid Network (FPN) and Path Aggregation Network (PAN) structures are used to input the output feature map of the feature extraction layer into the FPN-PAN network structure to achieve the fusion of multiple scale feature maps. FPN improves the detection capability of small targets by fusing features of different levels through a top-down path; PAN further adds a bottom-up path on this basis to achieve finer-grained feature aggregation, which not only reduces the amount of calculation, but also improves the feature fusion and context capture capabilities, and enhances the detection performance of targets of different sizes.

[0059] In an embodiment of the present invention, the head network (Head) layer of the network is responsible for target detection from the fused feature map, including two tasks: regression and classification. Among them, the design of the decoupled head (Decoupled-Head) is introduced in the Head layer, and the feature maps obtained by the network are respectively input into the decoupled head for prediction. In the regression branch, the position offset between the predicted box and the real box is calculated. In the classification branch, for each candidate box extracted from the anchor free box (Anchor Free), the region of interest pooling (RoIPooling) and convolution operations are performed to obtain a classifier output tensor. The value at each position represents the probability that the candidate box belongs to each category.

[0060] Finally, the final detection results are screened out through the Non-Maximum Suppression (NMS) method, the overlapping detection boxes are removed, and the most likely detection results are retained.

[0061] The training process of the embodiment of the present invention uses the YOLOv8 target detection model, in which the backbone network layer extracts rich features through convolution, deconvolution, residual connection and bottleneck structure, and uses the SPPF module for multi-scale pooling. The Neck layer adopts the FPN and PAN structures to achieve the fusion of multi-scale features and improve the detection ability and context capture of small targets. The Head layer introduces the Decoupled-Head design for predictive classification and regression, and the final detection results are screened out by the NMS method.

[0062] S240, model evaluation.

[0063] During the training process, the model is evaluated, and precision, recall, and average precision are used as the main criteria for evaluating the performance of the target detection algorithm. Precision refers to the proportion of true positive samples among all the cases judged as positive samples by the model; recall refers to the proportion of positive samples correctly identified by the model among all the actual positive samples. The formula is as follows:

[0064]

[0065]

[0066]

[0067] Among them, TP (True Positives) is the number of samples correctly identified as positive samples, FP (False Positives) is the number of negative samples mistakenly labeled as positive samples, FN (False Negatives) is the number of positive samples mistakenly labeled as negative samples, P is the precision, and R is the recall.

[0068] S250, model optimization.

[0069] Since the performance of the model during initial training fails to meet the requirements of high-precision detection, multiple trainings are required, and the training parameters and model structure must be adjusted to improve the performance of the model.

[0070] In an embodiment of the present invention, the process of model optimization is as follows:

[0071] Step S251, adjust the parameters of the model according to the performance of the model and the actual situation. If the convergence speed of the model is slow or fast, adjust the learning rate; if the overfitting phenomenon is serious after the model training, increase the batch size or reduce the number of iterations.

[0072] In some embodiments, in addition to adjusting the learning rate, batch size, and number of iterations according to the model convergence speed and overfitting, you can also try to use different learning rate adjustment strategies, such as exponential decay, cosine annealing, etc. In addition, the parameters of the batch normalization layer, the weight initialization method, etc. can also be optimized and adjusted to further improve the training effect of the model.

[0073] Step S252, using the stochastic gradient descent (SGD) optimizer to continuously iterate, improve the efficiency and stability of the training process, adjust the network weights and other parameters to minimize the loss function, use the complete intersection over union (CIoU) loss function to measure the loss of the rectangular box, and generate the optimal weight file. CIoU is a sophisticated and comprehensive loss function that helps to improve the positioning accuracy of the model in the target detection task. The calculation formula is as follows:

[0074]

[0075]

[0076]

[0077] LOSS CIoU =1-CIoU

[0078] In the CIoU loss function, w is the width of the prediction box, h is the height of the prediction box, and w gt is the width of the real box, h gt is the height of the real box. b and,b gt Represent the coordinates of the center of the predicted and true bounding boxes respectively. ρ is the Euclidean distance between the two center points, which is used to measure the offset of the spatial position. c is the length of the diagonal of the minimum closure area containing the two bounding boxes, which reflects the size of the bounding box. IoU represents the intersection ratio of the predicted and true bounding boxes, which measures their overlap. α and v take into account the aspect ratio of the bounding box, providing a comprehensive measurement to ensure that the loss function can more comprehensively evaluate the accuracy of the bounding boxes, including their location, size and shape, and ultimately achieve the effect of improving model performance.

[0079] In the optimization method of this method, the SGD optimizer and CIoU loss function are used to adjust the model parameters. Among them, the SGD optimizer effectively improves the training efficiency and stability, and the CIoU loss function further enhances the positioning accuracy of the model, thereby significantly improving the high-precision recognition performance, optimizing the model's detection ability for pine wood nematode disease, and solving the problem of insufficient model performance.

[0080] In addition, attention mechanisms can be further introduced into the model structure, such as channel attention mechanism, spatial attention mechanism or a hybrid attention mechanism combining the two. The attention mechanism can enable the model to pay more attention to the feature areas related to pine wood nematode disease when processing images, improve the efficiency and accuracy of feature extraction, and thus enhance the model's ability to identify diseases.

[0081] S260, model output. After multiple trainings and optimization of supplementary scene data, the model accuracy is 90.7%, the recall rate is 82.2%, and the average detection accuracy is 90%, which can achieve rapid and accurate identification of pine wood nematode disease from the perspective of drone inspections.

[0082] It should be noted that the training method of the above-mentioned insect disease detection model and the pine wood nematode disease identification method based on the perspective of drone inspection can be run in different electronic devices. For example, the insect disease detection model can be trained in a first electronic device and then sent to a second electronic device. When the second electronic device detects pine wood nematode disease, the trained insect disease detection model can be directly used to detect forest insect diseases. In addition, the training method and the detection method can be run in the same device. For example, after the drone inspection, the real original image taken is obtained, and the image is sent to an electronic device that stores the insect disease detection model. The electronic device runs the insect disease detection model to detect and judge the original image. The present application does not limit the operation mode of the execution subject.

[0083] The present application also provides an electronic device, comprising:

[0084] a memory for storing instructions to be executed by one or more processors of the device, and

[0085] A processor is used to execute the above embodiment in combination with Figures 1 to 5 The identification method explained.

[0086] The present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the processor executes the above-mentioned embodiment. Figures 1 to 6 The identification method explained.

[0087] The present application also provides a computer program product comprising instructions, which, when executed on an electronic device, enables a processor to execute the above-mentioned embodiments. Figures 1 to 5 The method explained.

[0088] Reference now Figure 6 , which is a block diagram of a SoC (System on Chip) 700 according to an embodiment of the present application. Figure 6In the FIG, similar components have the same reference numerals. In addition, the dashed boxes are optional features of more advanced SoCs. Figure 6 In the embodiment, SoC 700 includes: an interconnect unit 750, which is coupled to an application processor 710; a system agent unit 780; a bus controller unit 790; an integrated memory controller unit 740; a group or one or more coprocessors 720, which may include integrated graphics logic, an image processor, an audio processor, and a video processor; a static random access memory (SRAM) unit 730; and a direct memory access (DMA) unit 760. In one embodiment, the coprocessor 720 includes a dedicated processor, such as, for example, a network or communication processor, a compression engine, a general purpose computing on GPU (GPGPU), a high throughput integrated multi-core architecture (MIC) processor, or an embedded processor.

[0089] The static random access memory (SRAM) unit 730 may include one or more computer-readable media for storing data and / or instructions. The computer-readable storage medium may store instructions, specifically, temporary and permanent copies of the instructions. The instructions may include: when executed by at least one unit in the processor, the Soc 700 executes the method according to the above embodiment, which may be specifically referred to in the above embodiment. Figure 1 and Figure 5 The methods explained will not be repeated here.

[0090] The various embodiments of the mechanism disclosed in the present application can be implemented in hardware, software, firmware or a combination of these implementation methods. The embodiments of the present application can be implemented as a computer program or program code executed on a programmable system, which includes at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device and at least one output device.

[0091] Program code can be applied to input instructions to perform the functions described in this application and generate output information. The output information can be applied to one or more output devices in a known manner. For the purposes of this application, a processing system includes any system having a processor such as, for example, a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0092] Program code can be implemented with high-level programming language or object-oriented programming language to communicate with the processing system. When necessary, program code can also be implemented with assembly language or machine language. In fact, the mechanism described in this application is not limited to the scope of any specific programming language. In either case, the language can be a compiled language or an interpreted language.

[0093] In some cases, the disclosed embodiments may be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments may also be implemented as instructions carried or stored on one or more temporary or non-temporary machine-readable (e.g., computer-readable) storage media, which may be read and executed by one or more processors. For example, instructions may be distributed over a network or through other computer-readable media. Therefore, machine-readable media may include any mechanism for storing or transmitting information in a machine (e.g., computer) readable form, including, but not limited to, floppy disks, optical disks, optical disks, read-only memories (CD-ROMs), magneto-optical disks, read-only memories (ROMs), random access memories (RAMs), erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), magnetic or optical cards, flash memory, or a tangible machine-readable memory for transmitting information (e.g., carrier waves, infrared signals, digital signals, etc.) using the Internet in electrical, optical, acoustic, or other forms of propagation signals. Therefore, machine-readable media include any type of machine-readable media suitable for storing or transmitting electronic instructions or information in a machine (e.g., computer) readable form.

[0094] In the accompanying drawings, some structural or method features may be shown in a specific arrangement and / or order. However, it should be understood that such a specific arrangement and / or order may not be required. Instead, in some embodiments, these features may be arranged in a manner and / or order different from that shown in the illustrative drawings. In addition, the inclusion of structural or method features in a particular figure does not mean that such features are required in all embodiments, and in some embodiments, these features may not be included or may be combined with other features.

[0095] It should be noted that the units / modules mentioned in the various device embodiments of the present application are all logical units / modules. Physically, a logical unit / module can be a physical unit / module, or a part of a physical unit / module, or can be implemented as a combination of multiple physical units / modules. The physical implementation method of these logical units / modules themselves is not the most important. The combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed by the present application. In addition, in order to highlight the innovative part of the present application, the above-mentioned device embodiments of the present application do not introduce units / modules that are not closely related to solving the technical problems proposed by the present application, which does not mean that there are no other units / modules in the above-mentioned device embodiments.

[0096] It should be noted that in the examples and description of this patent, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one" do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0097] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle, characterized in that: include: Set the altitude, flight speed and shooting angle of the drone, and obtain images taken by the drone during inspection in the preset area; Inputting the image into an insect disease detection model to obtain an insect disease detection result; The insect disease detection model is trained by adjusting model parameters using a stochastic gradient descent optimizer and a complete intersection-over-union loss function.

2. The pine wood nematode disease identification method based on the inspection perspective of an unmanned aerial vehicle according to claim 1 is characterized in that: The training method of the insect disease detection model includes: Get the original image obtained from the inspection perspective of the drone; Preprocessing the original image to obtain an insect disease detection data set; The target detection model is trained based on the insect disease detection data. During the training process, the parameters of the target detection model are adjusted using a stochastic gradient descent optimizer and a complete intersection-over-union loss function, and the target detection model is trained to obtain the insect disease detection model.

3. The pine wood nematode disease identification method based on the inspection perspective of an unmanned aerial vehicle according to claim 2 is characterized in that: The preprocessing of the original image comprises: Cutting the original image of the original size into a plurality of image slices, removing similar and identical images; Each cut image is manually labeled to mark the location and category of pine wood nematode disease, thereby obtaining the pine disease detection data set.

4. The pine wood nematode disease identification method based on the inspection perspective of an unmanned aerial vehicle according to claim 3 is characterized in that: The preprocessing of the original image further includes: The annotated original image is enlarged, wherein the enlargement method includes at least one of cutting enhancement, random flipping, center cropping, random brightness transformation, random contrast transformation, random saturation transformation, adding Gaussian noise and mixed enhancement.

5. The method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to any one of claims 2 to 4, characterized in that: The target detection model includes: A backbone network, wherein the backbone network includes convolution, deconvolution, residual connection and bottleneck structure, and the backbone network uses a spatial pyramid pooling module to extract multi-scale features in the image; A neck network, used for fusing the extracted features at multiple scales; The head network,adopts the design of decoupled head to perform prediction,classification and regression on the features.

6. The method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to claim 5, characterized in that: The neck network adopts a feature pyramid network and a path aggregation network structure to fuse multiple scale feature maps.

7. The method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to claim 5, characterized in that: The training method of the insect disease detection model also includes: screening out the final detection result by a non-maximum suppression method, removing overlapping detection frames, and retaining the most likely detection result.

8. The method for identifying pine wood nematode disease based on the inspection perspective of an unmanned aerial vehicle according to claim 7, characterized in that: The training method of the insect disease detection model also includes: During the training of the target detection model, the trained model is evaluated, and the evaluation uses precision, recall, and average precision as the performance evaluation criteria for the target detection algorithm.

9. An electronic device, comprising: A memory for storing instructions executed by one or more processors of an electronic device, and a processor for executing the identification method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on an electronic device, enable the electronic device to perform the identification method according to any one of claims 1 to 8.

Images