A red kidney bean disease monitoring system based on the Internet of Things

Through the Internet of Things-based red kidney bean disease monitoring system, combined with the YOLOv5-SE-BiFPN target detection module, the BiFPN and attention mechanism are used to realize real-time monitoring and analysis of red kidney bean brown spot disease and bacterial epidemics, solving traditional monitoring methods The problem of subjectivity and low efficiency improves the accuracy and stability of disease detection.

CN118840667BActive Publication Date: 2024-11-29SHANXI AGRI UNIV
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202411333931.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-24
Publication Date
2024-11-29
Estimated Expiration
2044-09-24

AI Technical Summary

Technical Problem

The prior art is difficult to realize real-time monitoring and analysis of red kidney bean brown spot disease and bacterial epidemics, and traditional manual monitoring is highly subjective, time-consuming and inefficient.

Method used

The Internet of Things-based red kidney bean disease monitoring system is adopted, combined with the YOLOv5-SE-BiFPN target detection module, and the bidirectional characteristic pyramid network (BiFPN) and attention mechanism are introduced to monitor and analyze the disease status of red kidney beans in real time.

Benefits of technology

实现了对红芸豆褐斑病和细菌性疫病的实时监测和分析,提高了检测精度和稳定性,提供了详细的病害信息,帮助农业管理者做出科学决策,有效预防和控制病害的蔓延。

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118840667B_ABST
    Figure CN118840667B_ABST
Patent Text Reader

Abstract

The invention relates to the field of monitoring technology, and specifically discloses a red kidney bean disease monitoring system based on the Internet of Things. The system is used to solve the problem that no single leaf recognition data is further analyzed on the basis of red kidney bean brown spot disease and bacterial blight recognition to obtain more disease information about the red kidney bean brown spot disease and bacterial blight. The system comprises an image acquisition device, an image preprocessing module, a YOLOv5-SE-BiFPN target detection module, a brown spot disease and bacterial blight feature data extraction module, a brown spot disease and bacterial blight data analysis module and a data display module. By combining the Internet of Things technology and the YOLOv5-SE-BiFPN target detection module, the red kidney bean brown spot disease and bacterial blight are monitored in real time, the BiFPN and SE attention mechanisms are introduced, the detection accuracy and stability are improved, the disease spot feature data are extracted, the disease trend and severity are analyzed, the agricultural managers are assisted in making decisions, and the spread of the red kidney bean brown spot disease and bacterial blight is prevented and controlled.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of monitoring technology, and more specifically, to a red kidney bean disease monitoring system based on the Internet of Things. Background Art

[0002] Red kidney bean is an important economic crop with high medicinal and nutritional value. However, it is vulnerable to a variety of diseases and pests during its growth. Brown spot disease and bacterial blight are common diseases in the growth process of red kidney beans, which will cause the leaves of the plants to dry up and fall off, affect the pod setting of the crop, and seriously affect the yield of red kidney beans. With the rapid development of computer technology, the shortcomings of traditional manual monitoring of red kidney bean brown spot disease and bacterial blight, such as strong subjectivity, long time consumption, low efficiency and difficulty in achieving real-time monitoring, have been improved. In the public document "Improvement of the YOLOv5 Model in the Optimization of the Brown Spot Disease Recognition Algorithm of Kidney Bean", Pengyan Su et al. introduced a bidirectional feature pyramid network (BiFPN) and an attention mechanism into the YOLOv5 model to highlight the local features of red kidney bean brown spot and bacterial blight leaf images and filter key features, so that the model can more accurately identify the diseased areas of red kidney bean brown spot and bacterial blight while maintaining the detection speed of the original YOLOv5 model, improve the filtering ability of key features of red kidney bean brown spot and bacterial blight, and further optimize the model's detection performance for red kidney bean brown spot and bacterial blight. However, in order to eliminate the interference of the surrounding environment and other leaves, the model is targeted at The data collection time was uniformly stipulated as 11:00 a.m. to 2:00 p.m., and cropped images of single leaves affected by brown spot disease and bacterial blight were taken with a Canon 6D2 camera (26 million pixels) under sufficient outdoor natural light for identification. However, no further analysis of the single leaf identification data was performed on the basis of the identification of red kidney bean brown spot disease and bacterial blight to obtain more disease information about red kidney bean brown spot disease and bacterial blight. Real-time monitoring and analysis of brown spot disease and bacterial blight of red kidney beans in the field were carried out based on the Internet of Things, which effectively helped prevent the spread and aggravation of brown spot disease and bacterial blight of red kidney beans in the field. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, the present invention provides a red kidney bean disease monitoring system based on the Internet of Things, which combines the Internet of Things technology and the YOLOv5-SE-BiFPN target detection module to monitor red kidney bean brown spot disease and bacterial blight in real time, introduces BiFPN and attention mechanism, improves detection accuracy and stability, extracts disease spot feature data, analyzes disease trends and severity, assists agricultural managers in making decisions, and prevents and controls the spread of red kidney bean brown spot disease and bacterial blight.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A red kidney bean disease monitoring system based on the Internet of Things comprises an image acquisition device, an image preprocessing module, a YOLOv5-SE-BiFPN target detection module, a brown spot disease and bacterial disease feature data extraction module, a brown spot disease and bacterial disease data analysis module and a data display module, wherein the image acquisition device is connected to the image preprocessing module, the image preprocessing module is connected to the YOLOv5-SE-BiFPN target detection module, the YOLOv5-SE-BiFPN target detection module is connected to the brown spot disease and bacterial disease feature data extraction module, the brown spot disease and bacterial disease feature data extraction module is connected to the brown spot disease and bacterial disease data analysis module, the image preprocessing module, the YOLOv5-SE-BiFPN target detection module The PN target detection module, the brown spot disease and bacterial disease feature data extraction module, and the brown spot disease and bacterial disease data analysis module are all connected to the data display module. The image acquisition device is used to collect single leaf images of red kidney bean plants in the field and transmit the image data to the image preprocessing module. The growth of red kidney beans is monitored in real time based on the Internet of Things technology. The image preprocessing module is used to preprocess the collected images and transmit the processed images to the YOLOv5-SE-BiFPN target detection module and the data display module. The YOLOv5-SE-BiFPN target detection module introduces a bidirectional feature pyramid network (BiFPN) in the YOLOv5s target detection model for efficient bidirectional cross-scale connection and weighting Multi-scale feature fusion, introduces adaptively learned feature weights, performs top-down and bottom-up multi-scale feature fusion, compresses the spatial dimension of each feature map into a single value through global pooling to integrate the information in the channel, and weights each channel through the learned weights to highlight the information that is more critical to the network task. By re-weighting the original feature map with the learned channel weights, the network can focus more on specific features that are considered more important, and perform weighted calculations on the channels of each feature of red kidney bean brown spot and bacterial blight, optimize the expression of red kidney bean single leaf brown spot and bacterial blight features, accurately identify single leaves with red kidney bean brown spot and bacterial blight in the field image set, and identify red kidney bean brown spot. The single leaf image of spot disease and bacterial blight is transmitted to the brown spot disease and bacterial blight feature data extraction module and the data display module. The brown spot disease and bacterial blight feature data extraction module extracts the location information, spot area, spot number, spot perimeter and spot area of ​​the brown spot disease and bacterial blight spots from the identified single leaf image feature data of red kidney bean brown spot disease and bacterial blight, and transmits the data to the brown spot disease and bacterial blight feature analysis module and the data display module. The red kidney bean brown spot disease and bacterial blight extension trend analysis factor is obtained by using the preset brown spot disease and bacterial blight disease extension trend analysis formula, and the red kidney bean brown spot disease and bacterial blight disease severity score is obtained by using the preset brown spot disease and bacterial blight disease severity analysis formula.Then, the obtained red kidney bean brown spot disease and bacterial blight expansion trend analysis factor and the red kidney bean brown spot disease and bacterial blight disease severity score are multiplied to obtain the red kidney bean brown spot disease and bacterial blight prediction analysis value, and the obtained data is transmitted to the data display module.

[0006] As a further solution of the present invention, the YOLOv5-SE-BiFPN target detection module includes an Input end, a Backbone, a Neck and a Prediction. A single leaf image of red kidney beans is input at the Input end to extract features in the image. The spatial dimension of each feature map is compressed into a single value through global pooling to integrate the information in the channel. Each channel is weighted by the learned weight to highlight the information that is more critical to the network task. By re-weighting the learned channel weights with the original feature map, the network can focus more on specific features that are considered more important, and adaptively learn the weights of each channel of the features. At indexes 13 and 17, the Add operation in the bidirectional feature pyramid network (BiFPN) is used to fuse the P4 and P3 features in the Backbone to enhance the information interaction between the features. Finally, the model gradually transmits and processes the features of the image in the Neck, and the Prediction outputs the predicted location and category information of the spots in the single leaf image of red kidney beans.

[0007] The YOLOv5 model is a target detection model based on deep learning. Its main feature is that it significantly improves the detection speed while maintaining high detection accuracy. Therefore, it has become one of the preferred models in the field of target detection. However, when dealing with the recognition tasks of red kidney bean brown spot disease and bacterial blight, the model is difficult to accurately select and focus on the important features of red kidney bean brown spot disease and bacterial blight disease, and is prone to overfitting, which leads to a decrease in model recognition efficiency. The Bidirectional Feature Pyramid Network (BiFPN) is an efficient feature fusion network structure that is widely used in target detection and semantic segmentation in image processing technology and computer vision tasks. Through the bidirectional feature propagation mechanism, multi-layer features can be integrated and utilized to better capture the multi-scale features of the target. BiFPN adds contextual information to the edge of the original FPN module. The BiFPN network ignores nodes with only one input edge. The network believes that if a node has only one input edge, its contribution to the feature network is small because it does not have enough information for effective feature fusion, so its contribution to the network is small. In addition, additional edges are added between the input and output nodes in the same layer, and higher-level fusion features are obtained through iterative stacking. This strategy helps to improve the feature fusion efficiency and accuracy of the BiFPN network and enhance the model's feature extraction ability for specific scenarios. The BiFPN network aims to achieve effective fusion of input features with different resolutions and consider their contribution to the output features. First, the learning parameters are set to ensure the consistency of the input layer size and add additional weights to each feature layer. Secondly, the feature layers are weighted using a fast normalization fusion method. The development of the attention mechanism was inspired by the human visual system, which has shown excellent ability in processing complex images and extracting key features from complex backgrounds. Therefore, researchers applied this visual attention mechanism to machine vision, enabling the model to automatically select and focus on the parts of the input data that are more important to the task, thereby improving the performance of the model. Soft attention can iteratively calculate attention weights more conveniently in deep learning, so it is widely used in related research. The squeeze and excitation (SE) attention network extracts key features by weighting each feature channel. By adding the SE module to the YOLOv5x model, the model shows the highest performance in the identification of weeds and lettuce plants. The SE module effectively extracts the key features of tomato virus disease by weighting each feature channel, allowing the network to pay more attention to important feature channels while reducing attention to irrelevant information, thereby improving the model's ability to identify key features.

[0008] As a further solution of the present invention, in the YOLOv5-SE-BiFPN target detection module, the initial learning rate is 0.01, the momentum of the learning rate is 0.9, the batch size of each training is 8, the resolution of the input image is 640×640 pixels, the number of iterations is 300, and the performance detection evaluation indicators of the model are Precision, Recall and mAP. Precision represents the proportion of correctly detected defect frames, and mAP is a comprehensive evaluation indicator of the performance of the target detection model, which takes both Precision and Recall into consideration.

[0009] As a further solution of the present invention, in the image acquisition module, the field sampling area for red kidney bean brown spot disease and bacterial blight is determined by random sampling method, the image acquisition time is from 11:00 am to 14:00 pm, and the initial acquired image pixels are greater than or equal to 26 million pixels.

[0010] As a further solution of the present invention, in the image preprocessing module, Gaussian filtering is used to remove image noise, and histogram equalization is used to crop the image based on the automatic cropping algorithm of the region of interest, so that the image is adjusted to contain only a single leaf, and according to the characteristics of red kidney bean brown spot disease and bacterial blight, the spot area is manually located, and a rectangular border is drawn and named as the brown spot area.

[0011] As a further solution of the present invention, in the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease expansion trend analysis formula uses a logarithmic function to effectively process the current growth trend of the total area of ​​brown spot disease and bacterial disease. In the case of rapid initial growth, the rapidly changing data is smoothed, and the exponential function is used to process the change in the number of spots, and the change in the number of spots is captured in time. The trigonometric function is used to introduce a periodic change constant, and the influence of seasons on brown spot disease and bacterial disease is considered. The root value function is combined with the data characteristics of the total area and the total number of spots, and the Sigmoid function is used to standardize the final result value so that the output value is between 0 and 1. The preset spot disease expansion trend analysis formula is:

[0012] ;

[0013] Where: It is the expansion trend factor of brown spot disease and bacterial blight of red kidney bean. is the Sigmoid function, For the current moment, For time The total area of ​​lesions at the time, For time The number of lesions at a time, is a periodic constant, which is set according to the seasonal periodicity of brown spot disease and bacterial diseases. , , They are all weight adjustment factors, which are adjusted by fitting historical data.

[0014] As a further solution of the present invention, in the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease severity analysis formula uses the current total lesion area and the maximum total lesion area to normalize the current total lesion area, so that the area feature has the same scale for lesions of different sizes, takes the logarithm of the number of spots, reduces the impact of the difference in the number of spots on the results, smoothes the extreme changes in the number of spots, uses the current lesion perimeter and the maximum lesion perimeter to normalize the current lesion perimeter, ensures that the perimeter feature has an order of magnitude consistency between lesions of different perimeters, and uses the Sigmoid function to standardize the final result value. The preset brown spot disease and bacterial disease disease severity analysis formula is:

[0015] ;

[0016] Where: It is the disease severity analysis value of brown spot disease and bacterial blight of red kidney bean. For current time The total area of ​​lesions, is the maximum value of the total area of ​​lesions at the current time, is the number of lesions at the current moment, is the perimeter of the lesion at the current moment, is the maximum value of the lesion perimeter at the current moment, , , , is the weight adjustment factor, which is obtained through numerical analysis of historical data.

[0017] As a further solution of the present invention, the formula for predicting and analyzing the values ​​of red kidney bean brown spot and bacterial blight is:

[0018] ;

[0019] Where: It is the prediction and analysis value of red kidney bean brown spot disease and bacterial disease.

[0020] As a further solution of the present invention, the data display module is the monitoring terminal interface of the entire system, which displays the red kidney bean leaf images collected from the field and their preprocessing results in real time, updates and displays the disease images and features identified by the YOLOv5-SE-BiFPN target detection module in real time, displays various types of disease spot feature data extracted by the brown spot disease and bacterial blight feature data extraction module, including disease spot location information, disease spot area, number of spots, and disease spot circumference, displays the analysis results of the brown spot disease and bacterial blight data analysis module, including disease expansion trend factors, disease severity analysis values, and red kidney bean brown spot disease and bacterial blight prediction analysis values, provides a chart display function for historical data, facilitates users to compare the current disease situation with the historical situation, analyze the disease development trend, supports viewing the changes in disease data according to time periods, and automatically triggers an alarm when the disease expansion trend factor or the disease severity analysis value exceeds a preset threshold value to remind users to take corresponding prevention and control measures in time.

[0021] The technical effects and advantages of the present invention compared with the prior art are as follows: the present invention can monitor the development of red kidney bean brown spot disease and bacterial blight in real time by combining the Internet of Things technology and the YOLOv5-SE-BiFPN target detection module, making data collection and transmission more efficient, and realizing real-time monitoring and analysis of red kidney bean field diseases. The introduction of the bidirectional feature pyramid network (BiFPN) and the squeeze and excitation (SE) attention mechanism enables the YOLOv5-SE-BiFPN target detection module to more accurately identify red kidney beans while maintaining a high detection speed. The diseased areas of brown spot disease and bacterial blight are optimized by weighted calculation of each feature channel, and the expression of the characteristics of red kidney bean single leaf brown spot disease and bacterial blight is optimized, and the recognition accuracy and stability of the model are improved. The brown spot disease and bacterial blight feature data extraction module and the brown spot disease and bacterial blight data analysis module extract the feature data of the location information, area, number and circumference of the lesions, and use the preset formula to analyze the disease expansion trend and disease severity, providing comprehensive disease information, which can better assist agricultural managers in making decisions and effectively prevent and control the spread of diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a system block diagram of the present invention;

[0023] Figure 2 It is a network structure diagram of the YOLOv5-SE-BiFPN target detection module of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the present invention to clearly and completely describe the technical solutions in the present invention. Obviously, the technical solutions described are only part of the technical solutions of the present invention, not all of the technical solutions. Based on the content of the present invention, all other technical solutions obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] like Figure 1As shown, a red kidney bean disease monitoring system based on the Internet of Things proposed by the present invention includes an image acquisition device, an image preprocessing module, a YOLOv5-SE-BiFPN target detection module, a brown spot disease and bacterial disease feature data extraction module, a brown spot disease and bacterial disease data analysis module and a data display module, the image acquisition device is connected to the image preprocessing module, the image preprocessing module is connected to the YOLOv5-SE-BiFPN target detection module, the YOLOv5-SE-BiFPN target detection module is connected to the brown spot disease and bacterial disease feature data extraction module, the brown spot disease and bacterial disease feature data extraction module is connected to the brown spot disease and bacterial disease data analysis module, the image preprocessing module, the YOLOv5-SE-BiFPN target detection module -SE-BiFPN target detection module, brown spot and bacterial disease feature data extraction module, brown spot and bacterial disease data analysis module are all connected to the data display module. The image acquisition device is used to collect single leaf images of red kidney bean plants in the field and transmit the image data to the image preprocessing module. The growth of red kidney beans is monitored in real time based on the Internet of Things technology. The image preprocessing module is used to preprocess the collected images and transmit the processed images to the YOLOv5-SE-BiFPN target detection module and the data display module. The YOLOv5-SE-BiFPN target detection module introduces a bidirectional feature pyramid network (BiFPN) in the YOLOv5s target detection model for efficient bidirectional cross-scale connection. The network performs top-down and bottom-up multi-scale feature fusion by introducing adaptively learned feature weights, compressing the spatial dimension of each feature map into a single value through global pooling to integrate the information in the channel, and weighting each channel through the learned weights to highlight the information that is more critical to the network task. By re-weighting the original feature map with the learned channel weights, the network can focus more on specific features that are considered more important, and perform weighted calculations on the channels of each feature of red kidney bean brown spot and bacterial blight, optimize the expression of red kidney bean single leaf brown spot and bacterial blight features, accurately identify single leaves with red kidney bean brown spot and bacterial blight in the field image set, and identify the identified red kidney The single leaf image of bean brown spot disease and bacterial blight is transmitted to the brown spot disease and bacterial blight feature data extraction module and the data display module. The brown spot disease and bacterial blight feature data extraction module extracts the location information, spot area, spot number, spot perimeter and spot area of ​​the brown spot disease and bacterial blight spots from the identified single leaf image feature data of red kidney bean brown spot disease and bacterial blight, and transmits the data to the brown spot disease and bacterial blight feature analysis module and the data display module. The red kidney bean brown spot disease and bacterial blight extension trend analysis factor is obtained by using the preset brown spot disease and bacterial blight disease extension trend analysis formula, and the red kidney bean brown spot disease and bacterial blight disease severity score is obtained by using the preset brown spot disease and bacterial blight disease severity analysis formula.Then, the obtained red kidney bean brown spot disease and bacterial blight expansion trend analysis factor and the red kidney bean brown spot disease and bacterial blight disease severity score are multiplied to obtain the red kidney bean brown spot disease and bacterial blight prediction analysis value, and the obtained data is transmitted to the data display module.

[0026] Through the Internet of Things technology, the growth status and disease development of red kidney beans can be monitored in real time, problems can be discovered in time, and the risk of disease spread can be reduced. The automated image acquisition equipment and transmission module make data acquisition and transmission more efficient, reduce manual intervention, and improve work efficiency. The introduction of the bidirectional feature pyramid network (BiFPN) and attention mechanism optimizes the feature extraction and recognition capabilities of the YOLOv5 model to ensure that while maintaining a high detection speed, it can more accurately identify the areas affected by red kidney bean brown spot and bacterial blight. The BiFPN network realizes the efficient fusion of multi-scale features, improves the feature extraction capability of the model at different scales, and increases recognition accuracy and stability. The brown spot and bacterial blight feature data extraction module and data analysis The analysis module can extract and analyze various characteristic data of lesions, such as location information, area, quantity, and perimeter, and use preset formulas to analyze the disease expansion trend and severity, providing comprehensive disease information. Through detailed data analysis and disease prediction, it can help agricultural managers make scientific decisions, effectively prevent and control the spread of diseases, and reduce economic losses. When the disease expansion trend factor or severity analysis value exceeds the preset threshold, it can automatically trigger an alarm to remind users to take corresponding prevention and control measures in time, further improving the timeliness and effectiveness of disease management. The data display module provides a chart display function for historical data, which is convenient for users to compare the current disease situation with the historical situation, analyze the disease development trend, and formulate more effective management strategies.

[0027] like Figure 2 As shown in the figure, the YOLOv5-SE-BiFPN target detection module includes an Input end, Backbone, Neck and Prediction. A single leaf image of red kidney bean is input at the Input end to extract features from the image. The spatial dimension of each feature map is compressed into a single value through global pooling to integrate the information in the channel. Each channel is weighted by the learned weight to highlight the information that is more critical to the network task. By re-weighting the original feature map with the learned channel weight, the network can focus more on specific features that are considered more important, and adaptively learn the weight of each channel of the feature. At indexes 13 and 17, the Add operation in the bidirectional feature pyramid network (BiFPN) is used to fuse the P4 and P3 features in the Backbone to enhance the information interaction between the features. Finally, the model gradually transfers and processes the features of the image in the Neck, and the Prediction outputs the predicted location and category information of the spots in the single leaf image of red kidney bean.

[0028] Input an image at the input end, extract image features through Backbone, and use the SE attention mechanism to adaptively learn the weights of feature channels, so that the model pays more attention to important features and improves the accuracy of feature extraction. Use the Add operation in BiFPN to fuse the P4 and P3 features in Backbone at indexes 13 and 17. Use the bidirectional feature pyramid network (BiFPN) to achieve bidirectional cross-scale connection and weighted multi-scale feature fusion of features, enhance the information interaction between features, and enable the model to better capture the multi-scale features of the target. The attention mechanism adaptively learns the weights of feature channels, so that the model can automatically select and focus on the input data. More importantly, the model focuses more on key features and reduces the interference of irrelevant information. The features are weighted through the attention mechanism to optimize the expression of red kidney bean single leaf brown spot and bacterial blight features, improve the recognition accuracy and stability of the model, and reduce the occurrence of overfitting. The image features are transmitted and processed layer by layer in Neck. Through multiple fusions and processing, the expression ability of feature information is gradually enhanced, the detection performance of the model is improved, and the prediction results, including the location and category information of the spots, are output on the Prediction side. By accurately identifying the brown spot and bacterial blight spots in the red kidney bean single leaf images, detailed disease information is provided to help agricultural managers make scientific decisions.

[0029] The YOLOv5 model is a target detection model based on deep learning. Its main feature is that it significantly improves the detection speed while maintaining high detection accuracy. Therefore, it has become one of the preferred models in the field of target detection. However, when dealing with the recognition tasks of red kidney bean brown spot disease and bacterial blight, the model is difficult to accurately select and focus on the important features of red kidney bean brown spot disease and bacterial blight disease, and is prone to overfitting, which leads to a decrease in model recognition efficiency. The Bidirectional Feature Pyramid Network (BiFPN) is an efficient feature fusion network structure that is widely used in target detection and semantic segmentation in image processing technology and computer vision tasks. Through the bidirectional feature propagation mechanism, multi-layer features can be integrated and utilized to better capture the multi-scale features of the target. BiFPN adds contextual information to the edges of the original FPN module. Edge, the BiFPN network ignores nodes with only one input edge. The network believes that if a node has only one input edge, its contribution to the feature network is small because it does not have enough information for effective feature fusion and contributes less to the network. In addition, additional edges are added between the input and output nodes in the same layer, and higher-level fusion features are obtained through iterative stacking. This strategy helps to improve the feature fusion efficiency and accuracy of the BiFPN network and enhance the model's feature extraction ability for specific scenarios. The BiFPN network aims to achieve effective fusion of input features with different resolutions and consider their contribution to the output features. First, the learning parameters are set to ensure the consistency of the input layer size and add additional weights to each feature layer. Secondly, the feature layers are weighted using a fast normalization fusion method. The development of the attention mechanism was inspired by the human visual system, which has shown excellent ability in processing complex images and extracting key features from complex backgrounds. Therefore, researchers applied this visual attention mechanism to machine vision, enabling the model to automatically select and focus on the parts of the input data that are more important to the task, thereby improving the performance of the model. Soft attention can iteratively calculate attention weights more conveniently in deep learning, so it is widely used in related research. In the attention mechanism, key features are extracted by weighting each feature channel. By adding the SE module to the YOLOv5x model, the model shows the highest performance in the identification of weeds and lettuce plants. The SE module effectively extracts the key features of tomato virus disease by weighting each feature channel, allowing the network to pay more attention to important feature channels while reducing attention to irrelevant information, thereby improving the model's ability to identify key features.

[0030] It should be noted that in the YOLOv5-SE-BiFPN target detection module, the initial learning rate is 0.01, the momentum of the learning rate is 0.9, the batch size of each training is 8, the resolution of the input image is 640×640 pixels, the number of iterations is 300, and the performance detection evaluation indicators of the model are Precision, Recall and mAP. Precision represents the proportion of correctly detected defect boxes, and mAP is a comprehensive evaluation indicator of the target detection model performance, which takes both Precision and Recall into consideration.

[0031] The initial learning rate of 0.01 can ensure that the model converges quickly in the early stage of training, while avoiding training instability or divergence caused by too large a step size. Combined with momentum, the initial learning rate is gradually adjusted through the learning rate scheduling strategy to adapt to the needs of different training stages and improve model performance. The introduction of momentum accelerates model convergence and reduces the impact of gradient oscillation during training. The momentum mechanism can help the model jump out of the local optimal solution, making the training process smoother and more stable. The batch size of 8 reduces the memory usage to a certain extent, adapts to most GPU hardware conditions, and ensures the accuracy of gradient estimation. The moderate batch size helps to balance the training speed and model performance, making the training process efficient and the results reliable. Image resolution ensures that the model can capture more details and helps extract higher quality features. High-resolution image input improves detection accuracy, allowing the model to more accurately identify diseased areas. 300 iterations ensure that the model can fully learn the features in the data and avoid underfitting problems. A moderate number of iterations combined with other regularization measures can reduce the risk of overfitting and make the model more generalizable. Precision and Recall measure the accuracy and detection rate of the model respectively, while mAP takes both into account to provide a comprehensive performance evaluation. These evaluation indicators help determine the actual detection capability of the model and ensure that the model can reliably identify diseases and provide accurate detection results in actual applications.

[0032] It should be noted that in the image acquisition module, the field sampling areas for red kidney bean brown spot disease and bacterial blight were determined by random sampling. The image acquisition time was from 11:00 a.m. to 2:00 p.m., and the initial image pixels were greater than or equal to 26 million pixels.

[0033] The random sampling method can ensure that the sampling area is representative and covers plants in different positions and conditions throughout the field, thereby avoiding sampling bias. Through random sampling, the sampling points can be evenly distributed, reducing the systematic errors caused by human selection, and ensuring that the collected data are more universal and objective. The image acquisition time is from 11:00 am to 14:00 pm. During this period, the sun is sufficient and relatively stable, which can provide good natural lighting conditions, reduce the inconsistency of image quality caused by light changes, avoid the weak and drastic changes in light in the morning and evening, and ensure that the collected images have consistent brightness and contrast, which is conducive to subsequent image processing and analysis. The initial image pixels collected are greater than or equal to 26 million pixels, which can capture more detailed information, help to accurately identify and analyze the red kidney bean brown spot disease and bacterial blight lesions. In the subsequent processing and feature extraction process, multiple magnifications and detailed analysis can be performed without losing important feature information, thereby improving the accuracy of recognition.

[0034] It should be noted that in the image preprocessing module, Gaussian filtering is used to remove image noise, and histogram equalization is used to crop the image based on the automatic cropping algorithm of the region of interest, so that the image is adjusted to contain only a single leaf. According to the characteristics of red kidney bean brown spot disease and bacterial blight, the spot area is manually located, and a rectangular border is drawn and named as the brown spot area.

[0035] Gaussian filtering can effectively reduce the noise in the image, improve the stability of subsequent processing steps, and make the target area clearer and more identifiable. Histogram equalization can improve the contrast and brightness distribution of the image, making the image easier to observe and analyze, which is beneficial to subsequent feature extraction and detection. The automatic cropping algorithm can accurately identify the area of ​​interest, reduce the amount of data and time for processing, and improve the processing efficiency. Manually locating the spot area and drawing the rectangular border help to accurately locate the target area and provide clear marks for subsequent analysis, which is beneficial to the identification and analysis of brown spot disease and bacterial blight.

[0036] It should be noted that in the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease expansion trend analysis formula uses a logarithmic function to effectively process the current growth trend of the total area of ​​brown spot disease and bacterial disease. In the case of rapid growth in the initial stage, the sharply changing data is smoothed, and the exponential function is used to process the change in the number of spots, and the change in the number of spots is captured in time. The trigonometric function is used to introduce a periodic change constant, and the impact of seasons on brown spot disease and bacterial disease is considered. The root value function is combined with the data characteristics of the total area and the total number of spots, and the Sigmoid function is used to standardize the final result value so that the output value is between 0 and 1. The preset spot disease expansion trend analysis formula is:

[0037] ;

[0038] Where: It is the expansion trend factor of brown spot disease and bacterial blight of red kidney bean. is the Sigmoid function, For the current moment, For time The total area of ​​lesions at the time, For time The number of lesions at a time, is a periodic constant, which is set according to the seasonal periodicity of brown spot disease and bacterial diseases. , , They are all weight adjustment factors, which are adjusted by fitting historical data.

[0039] The use of logarithmic functions to smooth the current growth trend of the total area of ​​brown spot disease and bacterial disease can effectively eliminate noise and mutations in the data and more accurately reflect the development of the disease. The exponential function is used to process the changes in the number of spots, and the fluctuations in the number of spots are captured in time, which improves the sensitivity to changes in the number of diseases. The trigonometric function is introduced to process the periodic change constant, and the impact of seasons on brown spot disease and bacterial disease is considered, so that the analysis results are closer to the actual situation and the applicability and accuracy of the model are enhanced. The root value function is used to combine the data characteristics of the total area and the total number of spots, and the impact of the area and number of diseases on the expansion trend of the disease is comprehensively considered, making the analysis more comprehensive and accurate. The Sigmoid function is used to standardize the result values ​​and limit the output values ​​between 0 and 1, making the results easier to understand and compare, and improving the interpretability and operability of the analysis results.

[0040] It should be noted that in the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease severity analysis formula uses the current total lesion area and the maximum total lesion area to normalize the current total lesion area, so that the area feature has the same scale for lesions of different sizes, takes the logarithm of the number of spots, reduces the impact of the difference in the number of spots on the results, and smoothes the extreme changes in the number of spots. The current lesion perimeter and the maximum lesion perimeter are used to normalize the current lesion perimeter to ensure that the perimeter feature has an order of magnitude consistency between lesions of different perimeters. The Sigmoid function is used to standardize the final result value. The preset brown spot disease and bacterial disease disease severity analysis formula is:

[0041] ;

[0042] Where: It is the disease severity analysis value of brown spot disease and bacterial blight of red kidney bean. For current time The total area of ​​lesions, is the maximum value of the total area of ​​lesions at the current time, is the number of lesions at the current moment, is the perimeter of the lesion at the current moment, is the maximum value of the lesion perimeter at the current moment, , , , is the weight adjustment factor, which is obtained through numerical analysis of historical data.

[0043] The total area of ​​lesions, the number of lesions and the perimeter of lesions are normalized so that lesions of different sizes and shapes have the same scale, which is conducive to comprehensively considering the impact of these characteristics on the severity of the disease and improving the accuracy and comparability of the analysis. Taking the logarithm of the number of spots and smoothing the extreme changes in the number of spots reduces the impact of the difference in the number of spots on the results, making the analysis more stable and reliable and better reflecting the severity of the disease. The normalization process ensures that the perimeter characteristics of lesions have a consistent order of magnitude between lesions of different sizes, making the perimeter characteristics play a more stable and reliable role in the analysis. The Sigmoid function is used to standardize the final result value so that the result value range is between 0 and 1, which is easy to understand and compare, improving the interpretability and operability of the analysis results. The importance of different features is adjusted by the weight adjustment factor, which can flexibly control the analysis results according to actual conditions and enhance the applicability and adjustability of the model.

[0044] It should be noted that the formula for predicting and analyzing the values ​​of red kidney bean brown spot and bacterial blight is:

[0045] ;

[0046] Where: It is the prediction and analysis value of red kidney bean brown spot disease and bacterial disease.

[0047] This formula is used to calculate the predicted analysis value of red kidney bean brown spot disease and bacterial blight. Its mathematical meaning is to comprehensively evaluate the prediction of the disease based on the expansion trend and severity of red kidney bean brown spot disease and bacterial blight. Specifically: the numerator represents the comprehensive consideration of the expansion trend and severity of brown spot disease and bacterial blight, and the product of the two reflects the comprehensive impact of the disease; the denominator represents the sum of the expansion trend and severity, which is equivalent to the weighted sum of the two factors. Through such calculation, the obtained value reflects the comprehensive analysis value of the prediction of red kidney bean brown spot disease and bacterial blight. Taking into account the two key factors of the expansion trend and severity of brown spot disease and bacterial blight, the prediction of the disease is more comprehensively evaluated, which helps to improve the accuracy and credibility of the prediction. Through the product and sum of the numerator and denominator, the influence of the expansion trend and severity on the prediction results is balanced, making the prediction value more balanced and objective. Integrating the influence of different factors into a comprehensive indicator improves the interpretability of the results, making it easier for agricultural managers to understand and apply the prediction results, so as to take more targeted management measures.

[0048] It should be noted that the data display module is the monitoring terminal interface of the entire system. It displays the red kidney bean leaf images collected from the field and their preprocessing results in real time, updates and displays the disease images and features identified by the YOLOv5-SE-BiFPN target detection module in real time, displays various types of disease spot feature data extracted by the brown spot disease and bacterial blight feature data extraction module, including disease spot location information, disease spot area, number of spots, and lesion circumference, and displays the analysis results of the brown spot disease and bacterial blight data analysis module, including disease expansion trend factors, disease severity analysis values, and red kidney bean brown spot disease and bacterial blight prediction analysis values. It provides a chart display function for historical data, which is convenient for users to compare the current disease situation with the historical situation and analyze the disease development trend. It supports viewing the changes in disease data according to time periods. When the disease expansion trend factor or the disease severity analysis value exceeds the preset threshold, an alarm is automatically triggered to remind users to take corresponding prevention and control measures in time.

[0049] Users can view the red kidney bean leaf images, disease images and their characteristics collected in the field, as well as various disease spot characteristic data and data analysis results in real time, helping users to understand the health status of red kidney beans in a timely manner. It provides a chart display function for historical data, which is convenient for users to compare the current disease situation with the historical situation and analyze the development trend of the disease, so as to better formulate agricultural management strategies. When the disease expansion trend factor or the disease severity analysis value exceeds the preset threshold, the system can automatically trigger an alarm to remind users to take corresponding prevention and control measures in time, which helps to prevent the further spread and impact of the disease. It provides the function of viewing the changes in disease data by time period, allowing users to more conveniently obtain the required information, thereby more efficiently conducting agricultural management and decision-making.

[0050] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0051] Finally: The above description is only the preferred technical solution of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A red kidney bean disease monitoring system based on the Internet of Things, characterized in that: The invention comprises an image acquisition device, an image preprocessing module, a YOLOv5-SE-BiFPN target detection module, a brown spot disease and bacterial disease feature data extraction module, a brown spot disease and bacterial disease data analysis module and a data display module, wherein the image acquisition device is connected to the image preprocessing module, the image preprocessing module is connected to the YOLOv5-SE-BiFPN target detection module, the YOLOv5-SE-BiFPN target detection module is connected to the brown spot disease and bacterial disease feature data extraction module, the brown spot disease and bacterial disease feature data extraction module is connected to the brown spot disease and bacterial disease data analysis module, the image preprocessing module, the YOLOv5-SE-BiFPN target detection module, the brown spot disease and bacterial disease feature data extraction module, the brown spot disease and bacterial disease data analysis module, the image preprocessing module, the YOLOv5-SE-BiFPN target detection module, the brown spot disease and bacterial disease feature data extraction module, the brown spot disease and bacterial disease data analysis module, the The bacterial disease feature data extraction module, brown spot disease and bacterial disease data analysis module are all connected to the data display module. The image acquisition device is used to collect single leaf images of red kidney bean plants in the field and transmit the image data to the image preprocessing module. The growth of red kidney beans is monitored in real time based on the Internet of Things technology. The image preprocessing module is used to preprocess the collected images and transmit the processed images to the YOLOv5-SE-BiFPN target detection module and the data display module. The YOLOv5-SE-BiFPN target detection module introduces a bidirectional feature pyramid network (BiFPN) in the YOLOv5s target detection model for efficient bidirectional cross-scale connection and weighted multi-scale feature fusion. The adaptively learned feature weights are introduced to perform top-down and bottom-up multi-scale feature fusion. The spatial dimension of each feature map is compressed into a single value through global pooling to integrate the information in the channel. Each channel is weighted by the learned weights to highlight the information that is more critical to the network task. The original feature map is reweighted with the learned channel weights to enable the network to focus more on specific features that are considered more important. The channel of each feature of red kidney bean brown spot and bacterial blight is weighted, the expression of red kidney bean single leaf brown spot and bacterial blight features is optimized, and single leaves with red kidney bean brown spot and bacterial blight in the field image set are accurately identified, and the identified red kidney bean brown spot and bacterial blight are identified. The single leaf image of red kidney bean brown spot disease and bacterial blight is transmitted to the brown spot disease and bacterial blight feature data extraction module and the data display module. The brown spot disease and bacterial blight feature data extraction module extracts the location information, spot area, spot number, spot perimeter and spot area of ​​the brown spot disease and bacterial blight spots from the identified single leaf image feature data of red kidney bean brown spot disease and bacterial blight, and transmits the data to the brown spot disease and bacterial blight feature analysis module and the data display module. The red kidney bean brown spot disease and bacterial blight extension trend analysis factor is obtained by using the preset brown spot disease and bacterial blight disease extension trend analysis formula, and the red kidney bean brown spot disease and bacterial blight disease severity score is obtained by using the preset brown spot disease and bacterial blight disease severity analysis formula.Then, the obtained red kidney bean brown spot disease and bacterial blight expansion trend analysis factor and the red kidney bean brown spot disease and bacterial blight disease severity score are multiplied to obtain the red kidney bean brown spot disease and bacterial blight prediction analysis value, and the obtained data is transmitted to the data display module;, The YOLOv5-SE-BiFPN target detection module includes Input, Backbone, Neck and Prediction. A single leaf image of red kidney bean is input at the Input end to extract features from the image. The spatial dimension of each feature map is compressed into a single value through global pooling to integrate the information in the channel. Each channel is weighted by the learned weights to highlight the information that is more critical to the network task. The original feature map is reweighted with the learned channel weights so that the network can focus more on specific features that are considered more important. The weight of each channel of the feature is adaptively learned. At indexes 13 and 17, the Add operation in the bidirectional feature pyramid network (BiFPN) is used to fuse the P4 and P3 features in the Backbone to enhance the information interaction between features. Finally, the model gradually transfers and processes the features of the image in the Neck. The Prediction outputs the predicted location and category information of the spots in the single leaf image of red kidney bean. In the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease expansion trend analysis formula uses the logarithmic function to effectively process the current growth trend of the total area of ​​brown spot disease and bacterial disease. In the case of rapid growth in the initial stage, the rapidly changing data is smoothed. The exponential function is used to process the change in the number of spots, and the change in the number of spots is captured in time. The trigonometric function is used to introduce the periodic change constant, considering the impact of seasons on brown spot disease and bacterial disease. The root value function is used to combine the data characteristics of the total area and the total number of spots, and the Sigmoid function is used to standardize the final result value so that the output value is between 0 and 1. The preset spot disease expansion trend analysis formula is: Where: X sp is the expansion trend factor of brown spot disease and bacterial blight of red kidney bean, σ is the Sigmoid function, t is the current time, A(t+1) is the total area of ​​lesions at time t+1, N(t+1) is the number of lesions at time t+1, T is a periodic constant, which is set according to the seasonal periodicity of brown spot disease and bacterial blight, α, β, and γ are all weight adjustment factors, which are adjusted by fitting historical data; In the brown spot disease and bacterial disease data analysis module, the brown spot disease and bacterial disease disease severity analysis formula uses the current total lesion area and the maximum total lesion area to normalize the current total lesion area, so that the area feature has the same scale for lesions of different sizes, takes the logarithm of the number of spots, reduces the impact of the difference in the number of spots on the results, and smoothes the extreme changes in the number of spots. The current lesion perimeter and the maximum lesion perimeter are used to normalize the current lesion perimeter to ensure that the perimeter feature has an order of magnitude consistency between lesions of different perimeters. The Sigmoid function is used to standardize the final result value. The preset brown spot disease and bacterial disease disease severity analysis formula is: Where: X sd is the disease severity analysis value of brown spot disease and bacterial blight of red kidney bean, A(t) is the total area of ​​disease spots at the current time t, and A max (t) is the maximum total area of ​​the lesions at the current time, N(t) is the number of lesions at the current time, P(t) is the perimeter of the lesions at the current time, and P max (t) is the maximum value of the lesion perimeter at the current moment, λ1, λ2, λ3, and λ4 are weight adjustment factors obtained through numerical analysis of historical data; The formula for predicting and analyzing the value of red kidney bean brown spot and bacterial blight is: Where: X Z It is the prediction and analysis value of red kidney bean brown spot disease and bacterial disease.

2. The red kidney bean disease monitoring system based on the Internet of Things according to claim 1, characterized in that: In the YOLOv5-SE-BiFPN target detection module, the initial learning rate is 0.01, the momentum of the learning rate is 0.9, the batch size of each training is 8, the resolution of the input image is 640×640 pixels, the number of iterations is 300, and the performance detection evaluation indicators of the model are Precision, Recall and mAP.

3. The red kidney bean disease monitoring system based on the Internet of Things according to claim 1 is characterized in that: In the image acquisition module, the field sampling areas for red kidney bean brown spot disease and bacterial blight were determined by random sampling. The image acquisition time was from 11:00 a.m. to 2:00 p.m., and the initial image pixels were greater than or equal to 26 million pixels.

4. The red kidney bean disease monitoring system based on the Internet of Things according to claim 1, characterized in that: In the image preprocessing module, Gaussian filtering was used to remove image noise, and histogram equalization was used to crop the image based on the automatic cropping algorithm of the region of interest, so that the image was adjusted to contain only a single leaf. According to the characteristics of red kidney bean brown spot disease and bacterial blight, the spot area was manually located, and a rectangular border was drawn and named the brown spot area.

5. The red kidney bean disease monitoring system based on the Internet of Things according to claim 1 is characterized in that: The data display module is the monitoring terminal interface of the entire system. It displays the red kidney bean leaf images collected from the field and their preprocessing results in real time, updates and displays the disease images and features identified by the YOLOv5-SE-BiFPN target detection module in real time, displays various types of disease spot feature data extracted by the brown spot disease and bacterial blight feature data extraction module, including disease spot location information, disease spot area, number of spots, and lesion circumference, and displays the analysis results of the brown spot disease and bacterial blight data analysis module, including disease expansion trend factors, disease severity analysis values, and red kidney bean brown spot disease and bacterial blight prediction analysis values. It provides a chart display function for historical data, which is convenient for users to compare the current disease situation with the historical situation and analyze the disease development trend. It supports viewing the changes in disease data according to time periods. When the disease expansion trend factor and the disease severity analysis value exceed the preset threshold, an alarm is automatically triggered to remind users to take corresponding prevention and control measures in time.

Citation Information

Patent Citations

  • UV-b induced plant pathogen resistance

    CN113271764A

  • Mung bean leaf spot disease spot detection method and device based on LS-YOLOv5 network

    CN116030348A

  • Deep learning-based tomato disease and insect pest detection and identification method

    CN116630803A

  • Intelligent monitoring system for mung bean diseases and insect pests

    CN118366105A

  • Device and method for guiding in ultrasound assessment of an organ

    EP4144301A1