Rice seedling blast disease intelligent detection system based on deep learning

Abstract

The invention discloses a rice seedling blast disease intelligent detection system based on deep learning, relates to the technical field of crop detection, and solves the technical problems that early-stage diseases or diseases with atypical symptoms may not be accurately identified in time, further development of the diseases is caused, and greater loss is caused. Performing leaf and spike disease risk grading based on a feature extraction result, wherein different risk grades correspond to different prevention and control strategies; the system continuously monitors the image data of the growth area and the harvesting area, can timely find the subtle change of the characteristics of the leaves and the ears, and effectively reduces the yield loss and the quality reduction caused by diseases. Leaf and spike disease risk grading is combined with meteorological data of each management grid and hydrological data of an irrigation area; according to the system, meteorological data of each management grid and hydrological data of an irrigation area are combined, external disease condition risks are analyzed, and the external disease condition risks are combined with leaf and spike disease condition risk grading to carry out comprehensive risk early warning.

Description

Technical Field

[0001] The present invention belongs to the field of crop detection, and specifically is an intelligent detection system for rice seedling blast disease based on deep learning. Background Art

[0002] Rice blast is a major rice disease that can cause significant yield losses, ranging from 40% to 50% in severe cases to total crop failure. The disease occurs in various locations, most commonly affecting leaves and nodes, resulting in varying degrees of yield loss. Early and severe panicle and node blast, in particular, can cause white panicles and even total crop failure. Traditional disease diagnosis methods often rely on manual experience, which is susceptible to subjective factors and can lead to inaccurate identification of disease types. Farmers with different levels of experience may differ in their assessment of rice seedling blast symptoms, and subtle disease features may be overlooked, thus affecting diagnostic results.

[0003] Existing rice disease detection systems lack precise extraction and analysis of leaf and panicle characteristics, making it difficult to accurately distinguish diseases of different types and severity. They also lack multidimensional analysis combining meteorological and hydrological factors, which affects the timely and accurate identification of early-stage diseases or those with atypical symptoms, leading to further development of the disease and greater losses. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes an intelligent detection system for rice seedling blast disease based on deep learning, which is used to solve the technical problem that early-stage diseases or diseases with atypical symptoms may not be identified in a timely and accurate manner, leading to further development of the disease and causing greater losses.

[0005] To solve the above problems, the first aspect of the present invention provides a deep learning-based intelligent detection system for rice seedling blast disease, comprising: Data collection module: Obtain map data of the rice planting area, divide the rice planting area into harvesting area, growing area, irrigation area and auxiliary area, and set up management grids of different densities. Each management grid is equipped with a drone data collection module and a land data collection module; Regional image processing nodes: Based on the location of the management grid, regional image processing nodes are set up in the areas where several adjacent management grids are located. The regional image processing nodes receive data collected by the data acquisition modules in the surrounding management grids and obtain meteorological data of each management grid from the authorized meteorological center; Establish an image primary screening model at the regional image processing node to preliminarily classify images into risky images and safe images, and send the risky images and grid corresponding data to the cloud image processing platform; Cloud image processing platform: Extracts leaf morphological features and lesion features from image data of the growing area, and performs leaf disease risk classification; Extract ear morphological features and ear abnormality features from the image data of the harvested area, and classify the ear disease risk. Based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risks of each grid are analyzed, and combined with the leaf disease risk grading and ear disease risk grading, risk warnings are issued for regional grids.

[0006] Optionally, in an example of the above aspect, management grids of different densities are set, and each management grid is provided with a drone data collection module and a land data collection module, including the following steps: Set up a 100m×100m density grid in the harvesting area, a 200m×200m density grid in the growing area, and a hydrological monitoring module in the irrigation area; Each management grid is equipped with a drone data acquisition module and a land data acquisition module. The drone data acquisition module collects high-definition image data and infrared image data of rice leaves or ears; the land data acquisition module collects soil temperature and humidity, as well as high-definition image data and infrared image data of plant leaves.

[0007] Optionally, in an example of the above aspect, establishing a primary image screening model in a regional image processing node, preliminarily classifying images into risky images and safe images, and sending the risky images and grid corresponding data to a cloud image processing platform includes the following steps: Obtain historical data of multi-angle high-definition and infrared images of rice leaves or ears, label image data of rice leaves or ears at different stages in areas where seedling blast has occurred as risk images, and label image data of rice leaves or ears at different stages in areas where seedling blast has not occurred as safe images; By training the deep learning model with labeled images in the regional image processing node, an image primary screening model is established. The trained image primary screening model preliminarily classifies images into risky images and safe images. The safe images are retained locally, while the risk images and the grid's meteorological and soil data are sent to the cloud-based image processing platform.

[0008] Optionally, in an example of the above aspect, for the image data of the growth area, extracting leaf morphological features and lesion features from the image data, and performing leaf disease risk classification, includes the following steps: For the image data of the growth area, the pixel-level mask of the lesion is obtained through the DeepLabV3+ segmentation model. Combined with the area of the leaf detection frame, the proportion of the lesion area is calculated. By using the IoU threshold filtering, the noise area in the detection frame is excluded, and the aspect ratio, roundness and compactness feature data of the lesion are extracted for joint analysis of morphological features. The leaf disease risk is graded based on the leaf morphological characteristics.

[0009] Optionally, in an example of the above aspect, a pixel-level mask of the lesion is obtained by using a DeepLabV3+ segmentation model, and the lesion area ratio is calculated in combination with the area of the leaf detection frame, including the following steps: Collect high-definition images of rice leaves, including images of healthy leaves and leaves with varying degrees of seedling blast disease. Annotate the images, mark the leaf areas and diseased areas, and generate corresponding segmentation masks and detection frames. The processed image data is used to train DeepLabV3+ as a segmentation model, and the trained DeepLabV3+ model is applied to the image data collected from the growth area to generate a pixel-level mask of the lesion. Calculate the proportion of lesion area based on the segmentation mask:

[0010] Where Ar is the proportion of the lesion area, I(pi) is the indicator function, which is 1 if the predicted mask pixel value is greater than 0.5, otherwise it is 0, i∈(1,2,…,Nleaf); Nleaf is the total number of pixels in the leaf area.

[0011] Optionally, in an example of the above aspect, extracting characteristic data of aspect ratio, roundness and compactness of the lesions for joint analysis of morphological characteristics, and performing leaf disease risk classification in combination with leaf morphological characteristics, comprises the following steps: Extract the aspect ratio feature data of the lesion: extract the side length of the minimum circumscribed rectangle of the lesion area, and the aspect ratio feature value = the length of the short side of the rectangle / the length of the long side of the rectangle; Extract the roundness feature data of the lesion: roundness feature value = [4*π*(lesion area)] / (lesion area perimeter²); Extract the compactness feature data of the lesion: compactness feature value = lesion area / lesion area of the minimum circumscribed rectangle; In the historical data, after the lesions with an area smaller than the threshold area are detected in the grid, there is no lesion data that expands and spreads. The aspect ratio eigenvalue, roundness eigenvalue, and compactness eigenvalue of the corresponding lesion data are calculated, and the average of the corresponding eigenvalues is taken as the standard value; Calculate the degradation ratio of the aspect ratio characteristic value, roundness characteristic value and compactness characteristic value of the lesion data in the image data collected in the growth area compared with the standard value respectively; Obtain the percentage of diseased spots on leaves in the corresponding area of the grid, and rank the leaf disease risk based on the deterioration ratio of the aspect ratio, roundness, and compactness eigenvalues compared to the standard values: Grade=w1*Sf1+w2*Sf2+w3*Sf3+w4*Ar; Among them, Grade is the leaf disease risk grading coefficient, Sf1 is the deterioration ratio of the aspect ratio characteristic value compared to the standard value, Sf2 is the deterioration ratio of the roundness characteristic value compared to the standard value, Sf3 is the deterioration ratio of the compactness characteristic value compared to the standard value, w1, w2, w3 and w4 are weight coefficients; The leaf disease risk is graded according to the leaf disease risk grading coefficient.

[0012] Optionally, in an example of the above aspect, performing ear morphological feature extraction and ear abnormality feature extraction on the image data of the harvested area and grading ear disease risk includes the following steps: Based on the image data of the harvested area, a loss function for ear detection was constructed. The ear area was detected using the YOLOv8 target detection model, and the ear morphological features were extracted. The Mask R-CNN was combined to segment the ear structure, extract the ear length, ear density, and ear abnormal area features, and construct a joint ear-lesion evaluation function to grade the ear disease risk.

[0013] Optionally, in an example of the above aspect, constructing an ear detection loss function, detecting the ear region using a YOLOv8 object detection model, and extracting ear morphological features include the following steps: For the abnormal ear area, we construct an ear detection loss function and use the YOLOv8 target detection model to extract the ear area image:

[0014] Among them, L is the boundary regression accuracy of the extracted spike region image, IoU is the intersection over union ratio between the predicted box and the true box, is the Euclidean distance between the center of the predicted box and the center of the real box, c is the diagonal length of the minimum circumscribed rectangle between the predicted box and the real box, α is the weight coefficient, and v is the aspect ratio consistency coefficient. , r is the aspect ratio eigenvalue, C is the aspect ratio eigenvalue, K is the compactness eigenvalue, β1, β2 and β3 are the corresponding weights respectively; The morphological features of the ear are extracted, including the characteristic value of the length of the ear and the characteristic value of the curvature of the ear. The morphological features of the abnormal area of the ear are extracted, including the characteristic value of the length of the abnormal area and the texture analysis characteristic value. Among them, the texture analysis value is the characteristic value obtained by weighted average of four types of texture characteristic values: contrast, energy, homogeneity and correlation.

[0015] Optionally, in an example of the above aspect, constructing an ear-lesion joint evaluation function to grade ear disease risk includes the following steps: Construct the ear morphology evaluation function:

[0016] Among them, Lm is the ear morphology evaluation coefficient, Lsp is the characteristic value of ear length, Lsp0 is the standard value of ear length, Ksp is the characteristic value of ear curvature, Ksp0 is the standard value of ear curvature, are the corresponding weights of ear length and ear curvature; Construct the lesion feature evaluation function:

[0017] Among them, Le is the evaluation coefficient of the lesion characteristics of the ear, Lse is the characteristic value of the abnormal area length, Lse0 is the standard value of the abnormal area length, Kc is the characteristic value of the lesion texture analysis, Kc0 is the standard value of the lesion texture analysis, is the corresponding weight of abnormal area length and lesion texture analysis; According to the calculation formula of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient, the ear-lesion joint evaluation function was established:

[0018] Among them, Ljoi is the combined evaluation coefficient of ear and lesion, are the corresponding weights of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient; The ear disease risk is graded based on the ear-lesion joint assessment coefficient.

[0019] Optionally, in an example of the above aspect, based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risk of each grid is analyzed, and combined with the leaf disease risk classification and the ear disease risk classification, risk warning is issued for the regional grid, including the following steps: Obtain historical data on meteorological data for each management grid and hydrological data for irrigation areas when seedling plague occurred and when the region did not suffer from seedling plague, and mark seedling plague risk data and normal data; The spatiotemporal Transformer model is trained using the labeled data. The trained model identifies meteorological data from each management grid and hydrological data from irrigation areas as seedling blight risk data or normal data. If the leaf disease risk level or ear disease risk level of the grid reaches the most serious disease risk level, or the meteorological data of the management grid and the hydrological data of the irrigation area of the management grid are identified as seedling disease risk data, a risk warning will be issued for the corresponding grid.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention classifies leaf and ear disease risk based on feature extraction results, with different risk levels corresponding to different prevention and control strategies. The system continuously monitors image data from the growing and harvesting areas, and can promptly detect subtle changes in leaf and ear characteristics. Early warning prevents the further spread and spread of diseases, effectively reducing yield losses and quality declines caused by diseases. Taking rice seedling blast as an example, if not promptly controlled, it can lead to plant death and an increase in empty grains in the ear, seriously affecting rice yield and quality. Through early warning and intervention, losses can be kept to a minimum.

[0021] The present invention integrates meteorological data from each management grid and hydrological data from the irrigation area to stratify leaf and ear disease risk. The system analyzes external disease risk and combines this data with the leaf and ear disease risk stratification to generate comprehensive risk warnings. Meteorological factors such as temperature, humidity, and light, as well as hydrological data such as soil moisture and irrigation water volume, are closely related to the occurrence and progression of rice seedling blast. This multi-source data fusion analysis enables a more comprehensive and accurate assessment of disease risk, improving the reliability of early warnings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0023] Figure 1 Schematic diagram of the system framework of the present invention; Figure 2 This is a schematic diagram of the intelligent detection of rice seedling blast disease according to the present invention. DETAILED DESCRIPTION

[0024] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] See also Figure 1-Figure 2 The first embodiment of the present invention provides a rice seedling blast disease intelligent detection system based on deep learning, comprising: Data collection module: Obtain map data of the rice planting area, divide the rice planting area into harvesting area, growing area, irrigation area and auxiliary area, and set up management grids of different densities. Each management grid is equipped with a drone data collection module and a land data collection module; Regional image processing nodes: Based on the location of the management grid, regional image processing nodes are set up in the areas where several adjacent management grids are located. The regional image processing nodes receive data collected by the data acquisition modules in the surrounding management grids and obtain meteorological data of each management grid from the authorized meteorological center; Establish an image primary screening model at the regional image processing node to preliminarily classify images into risky images and safe images, and send the risky images and grid corresponding data to the cloud image processing platform; Cloud image processing platform: Extracts leaf morphological features and lesion features from image data of the growing area, and performs leaf disease risk classification; Extract ear morphological features and ear abnormality features from the image data of the harvested area, and classify the ear disease risk. Based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risks of each grid are analyzed, and combined with the leaf disease risk grading and ear disease risk grading, risk warnings are issued for regional grids.

[0026] In this embodiment, the rice planting area is subdivided into a harvesting area, a growing area, an irrigation area, and an auxiliary area. This refined regional division enables more targeted management of different functional areas.

[0027] Different densities of management grids can be flexibly adjusted based on the actual terrain, landforms, and planting density of the rice-growing area. In key monitoring areas or areas with complex terrain, denser grids can be set to obtain more detailed and accurate data.

[0028] Each management grid is equipped with both drone and land-based data collection modules, forming a three-dimensional data collection network. Drones can quickly scan large areas of rice cultivation from the air, acquiring macroscopic images and data, enabling the timely detection of large-scale disease outbreaks or abnormal growth.

[0029] Based on the location of management grids, regional image processing nodes are deployed in the areas of several adjacent management grids. This distributed data processing architecture effectively improves data processing efficiency. Each regional image processing node only needs to process data collected by the surrounding management grids, avoiding the network congestion and processing delays caused by the centralized transmission of large amounts of data to a central node for processing. Furthermore, multiple regional image processing nodes can operate in parallel, further accelerating data processing and enabling the system to analyze and diagnose collected data in a timely manner.

[0030] The distributed configuration of regional image processing nodes ensures excellent scalability. As the rice planting area expands or data collection demands increase, new regional image processing nodes can be easily added without requiring a major overhaul of the entire system. This distributed architecture also enhances the system's fault tolerance. If a regional image processing node fails, it only impacts the data processing of the surrounding management grids it is responsible for, without paralyzing the entire system. Other nodes continue to function normally, ensuring system stability and reliability.

[0031] Extract leaf morphological features and lesion features from the image data of the growing area, and perform leaf disease risk classification; Extract ear morphological features and ear abnormality features from the image data of the harvested area, and classify the ear disease risk. Based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risks of each grid are analyzed, and combined with the leaf disease risk grading and ear disease risk grading, risk warnings are issued for regional grids.

[0032] Extracting leaf morphology and lesion features from growing area images can accurately identify leaf changes caused by different types of diseases. For example, rice blast causes spindle-shaped lesions on leaves, and feature extraction can accurately capture this characteristic and distinguish it from other similar diseases. Extracting ear morphology and abnormal features from harvested areas can identify potential problems such as ear discoloration and deformities, providing a comprehensive understanding of rice health at all stages of growth.

[0033] Based on the feature extraction results, leaf and ear disease risk classification is performed, providing a scientific basis for disease prevention and control. Different risk levels correspond to different prevention and control strategies. For example, high-risk areas can be prioritized for personnel and resources, and high-efficiency pesticides can be used for targeted prevention and control, improving prevention and control effectiveness and reducing disease losses.

[0034] The system continuously monitors image data from growing and harvesting areas, promptly detecting subtle changes in leaf and ear characteristics. Early warning prevents the further spread of diseases, effectively reducing yield losses and quality declines caused by them. For example, rice seedling blast, if not promptly controlled, can lead to plant death and an increase in empty, barren grains in the ear, seriously impacting rice yield and quality. Through early warning and intervention, losses can be minimized.

[0035] The risk classification of leaf and ear disease conditions facilitates the rational allocation of human, material, and financial resources. Resources are concentrated on high-risk areas, while resources are appropriately reduced in low-risk areas. This risk warning mechanism enables farmers to conduct targeted field management, avoiding the time and energy wasted by comprehensive inspections and blind prevention efforts. Based on warning information, managers can quickly locate problem areas and implement appropriate management measures, improving management efficiency and effectiveness.

[0036] Leaf and ear disease risk grading integrates meteorological data from each management grid and hydrological data from the irrigation area. The system analyzes external disease risk conditions and combines this with leaf and ear disease risk grading to provide comprehensive risk warnings. Meteorological factors such as temperature, humidity, and light, as well as hydrological data such as soil moisture and irrigation water volume, are closely related to the occurrence and progression of rice seedling blast. This multi-source data fusion analysis enables a more comprehensive and accurate assessment of disease risk, improving the reliability of early warnings.

[0037] In one embodiment of the present invention, management grids of different densities are set, and each management grid is provided with a drone data acquisition module and a land data acquisition module, including the following steps: Set up a 100m×100m density grid in the harvesting area, a 200m×200m density grid in the growing area, and a hydrological monitoring module in the irrigation area; Each management grid is equipped with a drone data acquisition module and a land data acquisition module. The drone data acquisition module collects high-definition image data and infrared image data of rice leaves or ears; the land data acquisition module collects soil temperature and humidity, as well as high-definition image data and infrared image data of plant leaves.

[0038] In one embodiment of the present invention, a primary image screening model is established at a regional image processing node to preliminarily classify images into risky images and safe images, and the risky images and grid corresponding data are sent to a cloud image processing platform, including the following steps: Obtain historical data of multi-angle high-definition and infrared images of rice leaves or ears, label image data of rice leaves or ears at different stages in areas where seedling blast has occurred as risk images, and label image data of rice leaves or ears at different stages in areas where seedling blast has not occurred as safe images; By training the deep learning model with labeled images in the regional image processing node, an image primary screening model is established. The trained image primary screening model preliminarily classifies images into risky images and safe images. The safe images are retained locally, while the risk images and the grid's meteorological and soil data are sent to the cloud-based image processing platform.

[0039] In one embodiment of the present invention, leaf morphological feature extraction and lesion feature extraction are performed on the image data of the growth area, and leaf disease risk classification is performed, including the following steps: For the image data of the growing area, the pixel-level mask of the lesion is obtained through the DeepLabV3+ segmentation model. Combined with the area of the leaf detection frame, the proportion of the lesion area is calculated. By using the IoU threshold filtering, the noise area in the detection frame (such as the leaf edge and light reflection) is excluded, and the aspect ratio, roundness and compactness feature data of the lesion are extracted for joint analysis of morphological features. The leaf disease risk is graded based on the leaf morphological characteristics.

[0040] In one embodiment of the present invention, a pixel-level mask of the lesion is obtained by using the DeepLabV3+ segmentation model, and the lesion area ratio is calculated based on the area of the leaf detection frame, including the following steps: Collect high-definition images of rice leaves, including images of healthy leaves and leaves with varying degrees of seedling blast disease. Annotate the images, mark the leaf areas and diseased areas, and generate corresponding segmentation masks and detection frames. The processed image data is used to train DeepLabV3+ as a segmentation model, and the trained DeepLabV3+ model is applied to the image data collected from the growth area to generate a pixel-level mask of the lesion. Calculate the proportion of lesion area based on the segmentation mask:

[0041] Where Ar is the proportion of the lesion area, I(pi) is the indicator function, which is 1 if the predicted mask pixel value is greater than 0.5, otherwise it is 0, i∈(1,2,…,Nleaf); Nleaf is the total number of pixels in the leaf area.

[0042] In one embodiment of the present invention, the aspect ratio, roundness and compactness characteristic data of the lesions are extracted for joint analysis of morphological characteristics, and leaf disease risk classification is performed in combination with leaf morphological characteristics, including the following steps: Extract the aspect ratio feature data of the lesion: extract the side length of the minimum circumscribed rectangle of the lesion area, and the aspect ratio feature value = the length of the short side of the rectangle / the length of the long side of the rectangle; Extract the roundness feature data of the lesion: roundness feature value = [4*π*(lesion area)] / (lesion area perimeter²). The closer the roundness feature value is to 1, the closer the lesion shape is to a circle. Extract the compactness feature data of the lesion: compactness feature value = lesion area / lesion area of the minimum circumscribed rectangle. The closer the compactness feature value is to 1, the more compact the lesion shape is. In the historical data, after the lesions with an area smaller than the threshold of 5m×5m were detected in the grid, no lesion data of the lesion area was found to be expanding. The aspect ratio eigenvalue, roundness eigenvalue, and compactness eigenvalue of the corresponding lesion data were calculated, and the mean of the corresponding eigenvalues was taken as the standard value. Calculate the degradation ratio of the aspect ratio characteristic value, roundness characteristic value and compactness characteristic value of the lesion data in the image data collected in the growth area compared with the standard value respectively; After a large amount of experimental data statistics, it was found that the closer the lesion shape is to a circle, the more compact the lesion shape is, and the closer the aspect ratio is, the less likely the lesion is to spread. Therefore, when calculating the degradation ratio, the degradation ratio of the corresponding characteristic value compared to the standard value = (standard value - corresponding characteristic value) / standard value; Obtain the percentage of diseased spots on leaves in the corresponding area of the grid, and rank the leaf disease risk based on the deterioration ratio of the aspect ratio, roundness, and compactness eigenvalues compared to the standard values: Grade=w1*Sf1+w2*Sf2+w3*Sf3+w4*Ar; Among them, Grade is the leaf disease risk grading coefficient, Sf1 is the deterioration ratio of the aspect ratio characteristic value compared to the standard value, Sf2 is the deterioration ratio of the roundness characteristic value compared to the standard value, Sf3 is the deterioration ratio of the compactness characteristic value compared to the standard value, w1, w2, w3 and w4 are weight coefficients determined by experiments; The leaf disease risk is graded according to the leaf disease risk grading coefficient.

[0043] In this embodiment, the standard value calculated from the filtered data in the historical data is obtained, and the proportion of the diseased spot area in the corresponding region is obtained. The leaf disease risk grading coefficient is calculated using the leaf disease risk grading coefficient calculation formula, and the obtained value is used as the first leaf disease risk threshold; After detecting spots with an area larger than the threshold of 50m×50m in the historical data grid and with spots that have expanded and spread, calculate the aspect ratio eigenvalue, roundness eigenvalue, and compactness eigenvalue of the corresponding spot data, take the mean of the corresponding eigenvalues, and calculate the leaf disease risk grading coefficient. The obtained value is used as the second leaf disease risk threshold. When the leaf disease risk grading coefficient is greater than or equal to the second leaf disease risk threshold, it is determined to be a first-level leaf disease risk level; When the leaf disease risk grading coefficient is greater than or equal to the first leaf disease risk threshold and less than the second leaf disease risk threshold, it is determined to be a level three leaf disease risk level; When the leaf disease risk grading coefficient is less than the first leaf disease risk threshold, it is determined to be a level three leaf disease risk level.

[0044] In one embodiment of the present invention, for image data of a harvested area, extracting ear morphological features and ear abnormality features from the image data and grading ear disease risk includes the following steps: Based on the image data of the harvested area, a loss function for ear detection was constructed. The ear area was detected using the YOLOv8 target detection model, and the ear morphological features were extracted. The Mask R-CNN was combined to segment the ear structure, extract the ear length, ear density, and ear abnormal area features, and construct a joint ear-lesion evaluation function to grade the ear disease risk.

[0045] In one embodiment of the present invention, an ear detection loss function is constructed, the ear region is detected using a YOLOv8 object detection model, and ear morphological feature extraction is performed, including the following steps: For the abnormal ear area, we construct an ear detection loss function and use the YOLOv8 target detection model to extract the ear area image:

[0046] Among them, L is the boundary regression accuracy of the extracted spike region image, IoU is the intersection over union ratio between the predicted box and the true box, is the Euclidean distance between the center of the predicted box and the center of the real box, c is the diagonal length of the minimum circumscribed rectangle between the predicted box and the real box, α is the weight coefficient, α=v / [(1−IoU)+v], v is the aspect ratio consistency coefficient, , r is the aspect ratio eigenvalue, C is the aspect ratio eigenvalue, K is the compactness eigenvalue, β1, β2 and β3 are the corresponding weights respectively; The morphological features of the ear are extracted, including the characteristic value of the length of the ear and the characteristic value of the curvature of the ear. The morphological features of the abnormal area of the ear are extracted, including the characteristic value of the length of the abnormal area and the texture analysis characteristic value. Among them, the texture analysis value is the characteristic value obtained by weighted average of four types of texture characteristic values: contrast, energy, homogeneity and correlation.

[0047] In this embodiment, morphological feature extraction: The characteristic value of ear length = the diagonal length of the ear detection frame × pixel resolution;

[0048] Among them, (x1, y1) is the coordinate of the ear base, (x3, y3) is the coordinate of the ear tip, and (x2, y2) is the coordinate of the ear midpoint.

[0049] In one embodiment of the present invention, a combined ear-lesion evaluation function is constructed to grade ear disease risk, including the following steps: Construct the ear morphology evaluation function:

[0050] Among them, Lm is the ear morphology evaluation coefficient, Lsp is the characteristic value of ear length, Lsp0 is the standard value of ear length, Ksp is the characteristic value of ear curvature, Ksp0 is the standard value of ear curvature, are the corresponding weights of ear length and ear curvature; In this embodiment, the standard value of ear length and the standard value of ear curvature are obtained by collecting a large number of healthy ear images to detect the characteristic values of ear length and the characteristic values of ear curvature, and calculating the corresponding mean values, which are set as standard values.

[0051] Construct the lesion feature evaluation function:

[0052] Among them, Le is the evaluation coefficient of the lesion characteristics of the ear, Lse is the characteristic value of the abnormal area length, Lse0 is the standard value of the abnormal area length, Kc is the characteristic value of the lesion texture analysis, Kc0 is the standard value of the lesion texture analysis, is the corresponding weight of abnormal area length and lesion texture analysis; In this embodiment, the abnormal region length standard value and the texture analysis standard value are obtained by collecting a large amount of non-diffusion speckle image data to obtain abnormal region length feature values and texture analysis feature values, and calculating the corresponding mean values, which are set as standard values.

[0053] According to the calculation formula of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient, the ear-lesion joint evaluation function was established:

[0054] Among them, Ljoi is the combined evaluation coefficient of ear and lesion, are the corresponding weights of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient; The ear disease risk is graded based on the ear-lesion joint assessment coefficient.

[0055] In this embodiment, Set to 0.4 and 0.6, Set to 0.3 and 0.7, Set to 0.4 and 0.6.

[0056] When the historical data is obtained and each characteristic value is the standard value, the ear disease risk grading coefficient is 0, and 0 is used as the first ear disease risk threshold; After detecting lesions with an area larger than the threshold of 50m×50m in the historical data grid and with lesions expanding and spreading, the corresponding eigenvalues were calculated and the mean of the corresponding eigenvalues was taken to calculate the ear disease risk grading coefficient. The obtained value was used as the second ear disease risk threshold. When the ear disease risk grading coefficient is greater than or equal to the second ear disease risk threshold, it is determined to be the first ear disease risk level; When the ear disease risk grading coefficient is greater than or equal to the first ear disease risk threshold and less than the second ear disease risk threshold, it is determined to be the third ear disease risk level; When the ear disease risk grading coefficient is less than the first ear disease risk threshold, it is determined to be the third ear disease risk level.

[0057] In one embodiment of the present invention, based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risk of each grid is analyzed, and combined with the leaf disease risk classification and ear disease risk classification, risk warning is issued for the regional grid, including the following steps: Obtain historical data on meteorological data for each management grid and hydrological data for irrigation areas when seedling plague occurred and when the region did not suffer from seedling plague, and mark seedling plague risk data and normal data; The spatiotemporal Transformer model is trained using the labeled data. The trained model identifies meteorological data from each management grid and hydrological data from irrigation areas as seedling blight risk data or normal data. If the leaf disease risk level or ear disease risk level of the grid reaches the most serious disease risk level, or the meteorological data of the management grid and the hydrological data of the irrigation area of the management grid are identified as seedling disease risk data, a risk warning will be issued for the corresponding grid.

[0058] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A rice seedling blast disease intelligent detection system based on deep learning, characterized in that: include: Data collection module: Obtain map data of the rice planting area, divide the rice planting area into harvesting area, growing area, irrigation area and auxiliary area, and set up management grids of different densities. Each management grid is equipped with a drone data collection module and a land data collection module; Regional image processing nodes: Based on the location of the management grid, regional image processing nodes are set up in the areas where several adjacent management grids are located. The regional image processing nodes receive data collected by the data acquisition modules in the surrounding management grids and obtain meteorological data of each management grid from the authorized meteorological center; Establish an image primary screening model at the regional image processing node to preliminarily classify images into risky images and safe images, and send the risky images and grid corresponding data to the cloud image processing platform; Cloud image processing platform: Extracts leaf morphological features and lesion features from image data of the growing area, and performs leaf disease risk classification; Extract ear morphological features and ear abnormality features from the image data of the harvested area, and classify the ear disease risk. Based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risks of each grid are analyzed, and combined with the leaf disease risk grading and ear disease risk grading, risk warnings are issued for regional grids.

2. The deep learning-based intelligent detection system for rice seedling blast according to claim 1, characterized in that: Setting up management grids of different densities, and setting up a drone data acquisition module and a land data acquisition module in each management grid, including the following steps: Set up a 100m×100m density grid in the harvesting area, a 200m×200m density grid in the growing area, and a hydrological monitoring module in the irrigation area; Each management grid is equipped with a drone data acquisition module and a land data acquisition module. The drone data acquisition module collects high-definition image data and infrared image data of rice leaves or ears; the land data acquisition module collects soil temperature and humidity, as well as high-definition image data and infrared image data of plant leaves.

3. The rice seedling blast disease intelligent detection system based on deep learning according to claim 1, characterized in that: Establish an image primary screening model at the regional image processing node, preliminarily classify images into risky images and safe images, and send the risky images and grid corresponding data to the cloud image processing platform, including the following steps: Obtain historical data of multi-angle high-definition and infrared images of rice leaves or ears, label image data of rice leaves or ears at different stages in areas where seedling blast has occurred as risk images, and label image data of rice leaves or ears at different stages in areas where seedling blast has not occurred as safe images; By training the deep learning model with labeled images in the regional image processing node, an image primary screening model is established. The trained image primary screening model preliminarily classifies images into risky images and safe images. The safe images are retained locally, while the risk images and the grid's meteorological and soil data are sent to the cloud-based image processing platform.

4. The deep learning-based intelligent detection system for rice seedling blast according to claim 1, characterized in that: Based on the image data of the growth area, leaf morphological features and disease spot features are extracted from the image data, and leaf disease risk classification is performed, including the following steps: For the image data of the growth area, the pixel-level mask of the lesion is obtained through the DeepLabV3+ segmentation model. Combined with the area of the leaf detection frame, the proportion of the lesion area is calculated. By using the IoU threshold filtering, the noise area in the detection frame is excluded, and the aspect ratio, roundness and compactness feature data of the lesion are extracted for joint analysis of morphological features. The leaf disease risk is graded based on the leaf morphological characteristics.

5. The deep learning-based intelligent detection system for rice seedling blast according to claim 4, characterized in that: The DeepLabV3+ segmentation model is used to obtain the pixel-level mask of the lesion. Combined with the area of the leaf detection frame, the lesion area ratio is calculated, which includes the following steps: Collect high-definition images of rice leaves, including images of healthy leaves and leaves with varying degrees of seedling blast disease. Annotate the images, mark the leaf areas and diseased areas, and generate corresponding segmentation masks and detection frames. The processed image data is used to train DeepLabV3+ as a segmentation model, and the trained DeepLabV3+ model is applied to the image data collected from the growth area to generate a pixel-level mask of the lesion. Calculate the proportion of lesion area based on the segmentation mask: Where Ar is the proportion of the lesion area, I(pi) is the indicator function, which is 1 if the predicted mask pixel value is greater than 0.5, otherwise it is 0, i∈(1,2,…,Nleaf); Nleaf is the total number of pixels in the leaf area.

6. The deep learning-based intelligent detection system for rice seedling blast according to claim 4, characterized in that: Extract the aspect ratio, roundness and compactness characteristic data of the lesions for joint analysis of morphological characteristics, and combine the leaf morphological characteristics to perform leaf disease risk classification, including the following steps: Extract the aspect ratio feature data of the lesion: extract the side length of the minimum circumscribed rectangle of the lesion area, and the aspect ratio feature value = the length of the short side of the rectangle / the length of the long side of the rectangle; Extract the roundness feature data of the lesion: roundness feature value = [4*π*(lesion area)] / (lesion area perimeter²); Extract the compactness feature data of the lesion: compactness feature value = lesion area / lesion area of the minimum circumscribed rectangle; In the historical data, after the lesions with an area smaller than the threshold area are detected in the grid, there is no lesion data that expands and spreads. The aspect ratio eigenvalue, roundness eigenvalue, and compactness eigenvalue of the corresponding lesion data are calculated, and the average of the corresponding eigenvalues is taken as the standard value; Calculate the degradation ratio of the aspect ratio characteristic value, roundness characteristic value and compactness characteristic value of the lesion data in the image data collected in the growth area compared with the standard value respectively; Obtain the percentage of diseased spots on leaves in the corresponding area of the grid, and combine the deterioration ratio of the aspect ratio eigenvalue, roundness eigenvalue, and compactness eigenvalue compared to the standard value to perform leaf disease risk classification: Grade = w1*Sf1+w2*Sf2+w3*Sf3+w4*Ar; Among them, Grade is the leaf disease risk grading coefficient, Sf1 is the deterioration ratio of the aspect ratio characteristic value compared to the standard value, Sf2 is the deterioration ratio of the roundness characteristic value compared to the standard value, Sf3 is the deterioration ratio of the compactness characteristic value compared to the standard value, w1, w2, w3 and w4 are weight coefficients; The leaf disease risk is graded according to the leaf disease risk grading coefficient.

7. The deep learning-based intelligent detection system for rice seedling blast according to claim 1, characterized in that: Extracting ear morphological features and ear abnormality features from the image data of the harvested area and grading ear disease risk include the following steps: Based on the image data of the harvested area, a loss function for ear detection was constructed. The ear area was detected using the YOLOv8 target detection model, and the ear morphological features were extracted. The Mask R-CNN was combined to segment the ear structure, extract the ear length, ear density, and ear abnormal area features, and construct a joint ear-lesion evaluation function to grade the ear disease risk.

8. The deep learning-based intelligent detection system for rice seedling blast according to claim 7, characterized in that: Constructing the ear detection loss function, detecting the ear region using the YOLOv8 object detection model, and extracting the ear morphological features includes the following steps: For the abnormal ear area, we construct an ear detection loss function and use the YOLOv8 target detection model to extract the ear area image: Among them, L is the boundary regression accuracy of the extracted spike region image, IoU is the intersection over union ratio between the predicted box and the true box, is the Euclidean distance between the center of the predicted box and the center of the real box, c is the diagonal length of the minimum circumscribed rectangle between the predicted box and the real box, α is the weight coefficient, and v is the aspect ratio consistency coefficient. , r is the aspect ratio eigenvalue, C is the aspect ratio eigenvalue, K is the compactness eigenvalue, β1, β2 and β3 are the corresponding weights respectively; The morphological features of the ear are extracted, including the characteristic value of the length of the ear and the characteristic value of the curvature of the ear. The morphological features of the abnormal area of the ear are extracted, including the characteristic value of the length of the abnormal area and the texture analysis characteristic value. Among them, the texture analysis value is the characteristic value obtained by weighted average of four types of texture characteristic values: contrast, energy, homogeneity and correlation.

9. The deep learning-based intelligent detection system for rice seedling blast according to claim 7, characterized in that: Constructing a combined ear-lesion evaluation function to grade ear disease risk includes the following steps: Construct the ear morphology evaluation function: Among them, Lm is the ear morphology evaluation coefficient, Lsp is the characteristic value of ear length, Lsp0 is the standard value of ear length, Ksp is the characteristic value of ear curvature, Ksp0 is the standard value of ear curvature, are the corresponding weights of ear length and ear curvature; Construct the lesion feature evaluation function: Among them, Le is the evaluation coefficient of the lesion characteristics of the ear, Lse is the characteristic value of the abnormal area length, Lse0 is the standard value of the abnormal area length, Kc is the characteristic value of the lesion texture analysis, Kc0 is the standard value of the lesion texture analysis, is the corresponding weight of abnormal area length and lesion texture analysis; According to the calculation formula of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient, the ear-lesion joint evaluation function was established: Among them, Ljoi is the combined evaluation coefficient of ear and lesion, are the corresponding weights of the ear morphology evaluation coefficient and the ear lesion characteristic evaluation coefficient; The ear disease risk is graded based on the ear-lesion joint assessment coefficient.

10. The deep learning-based intelligent detection system for rice seedling blast according to claim 1, characterized in that: Based on the meteorological data of each management grid and the hydrological data of the irrigation area, the external disease condition risk of each grid is analyzed. Combined with the leaf disease risk classification and ear disease risk classification, risk warning is issued for the regional grid, including the following steps: Obtain historical data on meteorological data for each management grid and hydrological data for irrigation areas when seedling plague occurred and when the region did not suffer from seedling plague, and mark seedling plague risk data and normal data; The spatiotemporal Transformer model is trained using the labeled data. The trained model identifies meteorological data from each management grid and hydrological data from irrigation areas as either seedling blight risk data or normal data. If the leaf disease risk level or ear disease risk level of the grid reaches the most serious disease risk level, or the meteorological data of the management grid and the hydrological data of the irrigation area of the management grid are identified as seedling disease risk data, a risk warning will be issued for the corresponding grid.

Images