Real-time data rapid detection method for tomato leaf diseases

Abstract

The invention relates to the technical field of rapid detection of diseases, in particular to a tomato leaf disease real-time data rapid detection method which comprises the steps that a data set is generated based on a leaf image of tomato diseases, and the diseases comprise early blight, leaf mold and late blight of tomatoes; constructing a matrix expression model according to each leaf image feature in the data set, and performing feature filtering on the matrix expression model by using a volume sliding mode surface function; based on the filtered image feature model, feature re-extraction is carried out on the image feature model by using a normalization function; for the image feature model after feature re-extraction, calculating an observation error by using a model compensation observer; a control input vector of a volume sliding mode filtering fixed time is designed according to an observation error, and a tomato disease detection model is constructed in combination with a Lyapunov function, so that the model is higher in small target recognition rate and better in convergence, the tomato leaf recognition rate and detection speed are improved, and the tomato sampling leaf data image transmission and acquisition stability is improved.

Description

Technical Field

[0001] The invention relates to the technical field of rapid disease detection, and in particular to a method for rapid detection of tomato leaf diseases using real-time data. Background Art

[0002] Among vegetable crops worldwide, tomatoes play a vital role in vegetable cultivation. The global annual output reaches 170 million tons, and the significance of tomato leaves in the global vegetable market is obvious. Tomato leaves are essential to ensure the healthy growth of tomatoes. At present, the detection methods for tomato sampling leaves include combining partial convolution and embedded Gaussian non-local attention on the basis of YOLOv8, the two-stage target detection algorithm Fast R-CNN, and a lightweight YOLOv5 algorithm using multiple depth-separable convolutions.

[0003] After searching relevant patents and literature at home and abroad, there are references such as "Tomato detection method based on domain adaptive learning in dense planting environment" to solve the problem of poor adaptability of fruit detection model to different scenes and serious data dependence in dense planting environment. This study combines YOLOv5 model and domain adaptive learning to propose a tomato domain adaptive detection model TDA-YOLO, but this method mainly involves tomatoes themselves, and it is difficult to ensure rapid detection for thinner leaves. Another document "Lightweight tomato disease detection model in natural scenes" proposes a lightweight model based on improved YOLOv8 to solve the challenges in tomato disease detection in natural scenes, such as diverse disease information, different scales, leaf occlusion, etc., but this method is not effective in real-time detection under conditions of limited computing resources. In addition, the invention patent with patent application number: CN202410692399.8 "Application of SlyDCL2 gene in enhancing the resistance of tomatoes to tobacco mosaic virus" discloses the application of SlyDCL2 gene in enhancing the resistance of tomatoes to tobacco mosaic virus. The SlyDCL2 gene includes SlyDCL2a, SlyDCL2b, SlyDCL2c and SlyDCL2d. However, this method has high requirements for the viral immune system of tomatoes, which increases the difficulty of detection.

[0004] Although the above patents have some research on detection methods for tomato sampled leaves, these detection methods are difficult to achieve real-time image transmission effects for the leaves themselves, and due to the complex dense planting environment, a real-time data rapid detection method for tomato leaf diseases is proposed through a volumetric sliding mode method and a model compensation observer. It has the ability to adjust weights online and filter data of tomato sampled leaves to prevent the unreasonable situation that non-tomato leaf diseases are identified as diseases, and does not need to rely too much on digital image processing and traditional machine learning technology, which increases the requirements for real-time detection of tomato sampled leaves and protects tomatoes from pests and pathogens.

[0005] Aiming at the shortcomings of the existing technology, in order to solve the problems of poor adaptability of tomato leaf detection model to the external environment and serious model dependence in densely planted environment, a rapid detection method for tomato leaf disease data was proposed by volumetric sliding mode method and model compensation observer. Summary of the invention

[0006] The object of the present invention is to provide a rapid detection method for tomato leaf diseases based on real-time data to solve the problems raised in the above background technology.

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

[0008] A method for rapid detection of tomato leaf disease real-time data, the method comprising:

[0009] Generate a dataset based on leaf images of tomato diseases, including early blight, leaf mold and late blight of tomato;

[0010] A matrix expression model is constructed according to the features of each leaf image in the data set, and feature filtering is performed on the matrix expression model using a volumetric sliding surface function;

[0011] Based on the filtered image feature model, a normalization function is used to re-extract features from the image feature model;

[0012] The observation error is calculated using the model compensation observer for the image feature model after feature extraction;

[0013] The fixed-time control input vector of the volumetric sliding mode filter is designed according to the observation error, and a tomato disease detection model is constructed in combination with the Lyapunov function.

[0014] Preferably, a dataset is generated based on leaf images of tomato diseases, including:

[0015] Identify common tomato disease characteristics based on the densely planted tomato growing environment;

[0016] Collect leaf images of tomato diseases under common disease characteristics and use the image feature extraction formula:

[0017]

[0018] Extract feature information of each leaf image to obtain a data set;

[0019] Among them, θ n is the sampling feature layer, L is the number of output feature layers, n and m are both positive numbers, M(θ n ) is the output feature layer inertia matrix, H(η n) is the output feature layer restoring force matrix, and P is the output feature layer index of the data set.

[0020] Preferably, constructing a matrix expression model according to the image features of each leaf in the data set includes:

[0021] Obtain the position resolution of each leaf image in the data set, and set the position resolution of the leaf image to two variables α and k from high to low;

[0022] Get the local features of each leaf image in the data set, and use the local feature detection algorithm to generate a matrix expression model:

[0023]

[0024] Among them, y i is the pose state vector after the global optimal solution operation, which is used to represent the matrix expression model, ω c is the actual value of the local operator pose space feature, is the local operator pose space feature estimate, is the normalization function, x c is the total perturbation value of the image in the dataset, is the total perturbation estimate of the image in the data set, α is a positive number, and k is a constant.

[0025] Preferably, the feature filtering of the matrix expression model is performed using a volumetric sliding surface function, including:

[0026] Get the matrix expression model y i , according to the formula: Get the filtered image feature g(y i );

[0027] Among them, W g represents the training weight matrix parameter, P represents the output feature layer index, ξ1 and ξ2 represent bilateral filter constants, and x c is the total perturbation value of the image in the dataset, is the estimated value of the total perturbation of the image in the dataset.

[0028] Preferably, re-extracting features from the image feature model using a normalization function includes:

[0029] Based on the filtered image features g(y i ), using non-local mean and bilateral filtering methods to filter out information with overlapping and chaotic features in the image features;

[0030] And use the normalization function to simplify the filtered image features to obtain a new image feature model

[0031]

[0032] in, is the estimated value of the total perturbation of the image in the dataset, is the local operator pose spatial feature estimation value, and β is the feature map of tomato sampled leaves at different scales.

[0033] Preferably, the observation error is calculated using a model compensation observer for the image feature model after feature re-extraction, including:

[0034] Get the actual value ω of the local operator pose space feature in the matrix expression model c and local operator pose space feature estimates Calculate ω separately c and The distance between the two and the diagonal distance, and ω c and The distance between them is recorded as D, and the diagonal distance is recorded as

[0035] Get a new image feature model Computing Observation Errors Using Model Compensated Observers

[0036]

[0037] Among them, α represents the enhanced feature layer of the tomato sampled leaves, v represents filtering smoothing, N represents the correction factor, and sig(g) represents the Gaussian radial basis function.

[0038] Preferably, a control input vector of a fixed time of a volumetric sliding mode filter is designed according to the observation error, and a tomato disease detection model is constructed in combination with a Lyapunov function, including:

[0039] Extract leaf image features that classify tomato leaves into green ripe stage, color change stage and red ripe stage in the data set respectively, confirm that the output feature layer index of the green ripe stage leaf image is w, the output feature layer index of the color change stage leaf image is ρ, and the output feature layer index of the red ripe stage leaf image is h;

[0040] Get the observation error The gamma function is used to obtain the fixed-time control input vector χ of the cubature sliding mode filter:

[0041]

[0042] Where D1 and D2 represent the Euclidean distance between the actual value of the local operator pose space feature and the estimated value of the local operator pose space feature, respectively;

[0043] The tomato disease detection model V(χ) is generated using the Lyapunov function:

[0044]

[0045] Among them, μ(χ) represents the selection accuracy parameter, θ(χ) represents the recall function, represents the average precision mean measurement function, g(y i ) represents the filtered image features, and i and n both represent integers.

[0046] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0047] Compared with the adaptive detection model TDA-YOLO method, a rapid disease data detection method implemented by the volumetric sliding mode method and the model compensation observer based on the present invention can estimate the external environmental interference within a fixed time, and can effectively compensate for the difference between the source domain and the target domain during the data image transmission process, and has a better recognition effect, so that the image transmission model has a higher recognition rate and better convergence for small targets, improves the recognition rate and detection speed of tomato leaves, and improves the stability of image transmission acquisition of tomato sampling leaf data.

[0048] Compared with the convolutional neural network method, this method is a rapid disease data detection method implemented by the volumetric sliding mode method and the model compensation observer. It has the ability to adjust weights online and filter data of tomato sampled leaves to prevent the unreasonable situation that non-tomato leaf diseases are identified as diseases. It does not have to rely too much on digital image processing and traditional machine learning technology, and increases the requirement for real-time detection of tomato sampled leaves to protect tomatoes from pests and pathogens.

[0049] The present invention can estimate the external environmental interference within a fixed time, can effectively compensate for the difference between the source domain and the target domain in the process of data image transmission, and has a better recognition effect, so that the image transmission model has a higher recognition rate and better convergence for small targets, improves the recognition rate and detection speed of tomato leaves, and improves the stability of tomato sampling leaf data image transmission acquisition. It has the ability to adjust weights and filter data of tomato sampling leaves online to prevent the unreasonable situation that non-tomato leaf diseases are identified as diseases, and does not need to rely too much on digital image processing and traditional machine learning technology, increases the requirement for real-time detection of tomato sampling leaves, and protects tomatoes from pests and pathogens. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0051] Figure 1It is a flow chart of a method for rapid detection of tomato leaf disease real-time data of the present invention;

[0052] Figure 2 This is a comparison chart of the effects of the method of the present invention and the tomato domain adaptive detection model TDA-YOLO method. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-Figure 2 , the present invention provides a technical solution:

[0055] A method for rapid detection of tomato leaf disease real-time data, the method comprising:

[0056] Generate a dataset based on leaf images of tomato diseases, including early blight, leaf mold and late blight of tomato;

[0057] Preferably, a dataset is generated based on leaf images of tomato diseases, including:

[0058] Confirming common disease characteristics of tomatoes according to the densely planted growth environment of tomatoes; wherein the common disease characteristics include leaf disease characteristics of early blight, leaf mold and late blight of tomatoes;

[0059] Collect leaf images of tomato diseases under common disease characteristics and use the image feature extraction formula:

[0060]

[0061] Extract feature information of each leaf image to obtain a data set;

[0062] Among them, θ n is the sampling feature layer, L is the number of output feature layers, n and m are both positive numbers, M(θ n ) is the output feature layer inertia matrix, H(η n ) is the output feature layer restoring force matrix, and P is the output feature layer index of the data set.

[0063] A matrix expression model is constructed according to the features of each leaf image in the data set, and feature filtering is performed on the matrix expression model using a volumetric sliding surface function;

[0064] Preferably, constructing a matrix expression model according to the image features of each leaf in the data set includes:

[0065] Obtain the position resolution of each leaf image in the data set, and set the position resolution of the leaf image to two variables α and k from high to low;

[0066] Get the local features of each leaf image in the data set, and use the local feature detection algorithm to generate a matrix expression model:

[0067]

[0068] Among them, y i is the pose state vector after the global optimal solution operation, which is used to represent the matrix expression model, ω c is the actual value of the local operator pose space feature, which is used to represent the actual position of the leaf image in the two-dimensional space; is the local operator pose space feature estimate, is a normalization function, which is used to ensure that the overall information before and after the transformation is the same; x c is the total perturbation value of the image in the dataset, is the total perturbation estimate of the image in the data set, α is a positive number, and k is a constant;

[0069] By adjusting the two variables α and k in the calculation position resolution from high to low, the measurement error of the tomato diseased leaves can be converged to the expected similarity within a fixed time, so as to obtain the leaves most likely to be tomato diseased.

[0070] Preferably, the feature filtering of the matrix expression model is performed using a volumetric sliding surface function, including:

[0071] Get the matrix expression model y i , according to the formula: Get the filtered image feature g(y i );

[0072] The volumetric sliding mode surface function is obtained by analyzing the volumetric Kalman filter and the sliding mode control method;

[0073] Among them, W g represents the training weight matrix parameter, P represents the output feature layer index, ξ1 and ξ2 represent bilateral filter constants, and x c is the total perturbation value of the image in the dataset, is the estimated value of the total perturbation of the image in the dataset.

[0074] Based on the filtered image feature model, a normalization function is used to re-extract features from the image feature model;

[0075] Preferably, re-extracting features from the image feature model using a normalization function includes:

[0076] Based on the filtered image features g(yi ), since the sampled tomato leaves may contain repeated and useless features, the non-local mean and bilateral filtering methods are used to filter out the information of feature overlap and chaotic features in the image features;

[0077] And use the normalization function to simplify the filtered image features to obtain a new image feature model The extracted image features have removed the information of the above-mentioned overlapping and confusing features:

[0078]

[0079] in, is the total perturbation estimate of the image in the dataset, is the local operator pose spatial feature estimation value, and β is the feature map of tomato sampled leaves at different scales.

[0080] The observation error is calculated using the model compensation observer for the image feature model after feature extraction;

[0081] Preferably, the observation error is calculated using a model compensation observer for the image feature model after feature re-extraction, including:

[0082] Get the actual value ω of the local operator pose space feature in the matrix expression model c and local operator pose space feature estimates Calculate ω separately c and The distance between and diagonal distance between c and The distance between them is recorded as D, and the diagonal distance is recorded as

[0083] Get a new image feature model Computing Observation Errors Using Model Compensated Observers

[0084]

[0085] Among them, α represents the enhanced feature layer of tomato sampled leaves, v represents filter smoothing, N represents the correction factor, and sig(g) represents the Gaussian radial basis function;

[0086] In the above formula, when the model compensation observer accurately estimates the external disturbance δ, then the observation error It can converge to a neighborhood near the origin in a fixed time;

[0087] By using the model compensation observer to calculate the observation error, the similarity distance between the current position of the tomato sampling leaf and the positions of other tomato leaves is calculated, the contribution of all other positions to the current position is obtained, and the error is confirmed, which is conducive to providing a better identification effect.

[0088] The control input vector of the fixed time of the volumetric sliding mode filter is designed according to the observation error, and the tomato disease detection model is constructed by combining the Lyapunov function.

[0089] Preferably, a control input vector of a fixed time of a volumetric sliding mode filter is designed according to the observation error, and a tomato disease detection model is constructed in combination with a Lyapunov function, including:

[0090] Extract leaf image features that classify tomato leaves into green ripe stage, color change stage and red ripe stage in the data set respectively, confirm that the output feature layer index of the green ripe stage leaf image is w, the output feature layer index of the color change stage leaf image is ρ, and the output feature layer index of the red ripe stage leaf image is h;

[0091] Get the observation error Considering factors such as distortion and feature overlap of tomato leaf sampling images, there will be problems such as reference position and information lag in long-distance transmission after the fusion of multiple sets of image features. The gamma function is used to obtain the fixed-time control input vector χ of the volumetric sliding mode filter:

[0092]

[0093] Where D1 and D2 represent the Euclidean distance between the actual value of the local operator pose space feature and the estimated value of the local operator pose space feature, respectively;

[0094] The tomato disease detection model V(χ) is generated using the Lyapunov function:

[0095]

[0096] Among them, μ(χ) represents the selection accuracy parameter, θ(χ) represents the recall function, represents the average precision mean measurement function, g(y i ) represents the filtered image features, and i and n both represent integers;

[0097] In the formula, V(χ) is derived, and the accuracy parameter μ(χ), the recall function θ(χ), and the average precision mean measurement function are selected. The detection speed of the tomato leaf sampling transmission data model is made to complete the minimum convergence within a fixed time, thereby improving the recognition rate and detection speed of tomato leaves.

[0098] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for rapid detection of tomato leaf diseases in real time, characterized in that: The method comprises: Generate a dataset based on leaf images of tomato diseases, including early blight, leaf mold and late blight of tomato; A matrix expression model is constructed according to the features of each leaf image in the data set, and feature filtering is performed on the matrix expression model using a volumetric sliding surface function; Based on the filtered image feature model, a normalization function is used to re-extract features from the image feature model; The observation error is calculated using the model compensation observer for the image feature model after feature extraction; The fixed-time control input vector of the volumetric sliding mode filter is designed according to the observation error, and a tomato disease detection model is constructed in combination with the Lyapunov function.

2. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The leaf image generation data set based on tomato diseases includes: Identify common tomato disease characteristics based on the densely planted tomato growing environment; Collect leaf images of tomato diseases under common disease characteristics and use the image feature extraction formula: Extract feature information of each leaf image to obtain a data set; Among them, θ n is the sampling feature layer, L is the number of output feature layers, n and m are both positive numbers, M(θ n ) is the output feature layer inertia matrix, H(η n ) is the output feature layer restoring force matrix, and P is the output feature layer index of the data set.

3. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The step of constructing a matrix expression model according to the image features of each leaf in the data set includes: Obtain the position resolution of each leaf image in the data set, and set the position resolution of the leaf image to two variables α and k from high to low; Get the local features of each leaf image in the data set, and use the local feature detection algorithm to generate a matrix expression model: Among them, y i is the pose state vector after the global optimal solution operation, which is used to represent the matrix expression model, ω c is the actual value of the local operator pose space feature, is the local operator pose space feature estimate, is the normalization function, x c is the total perturbation value of the image in the dataset, is the total perturbation estimate of the image in the data set, α is a positive number, and k is a constant.

4. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The step of using the volumetric sliding surface function to perform feature filtering on the matrix expression model includes: Get the matrix expression model y i , according to the formula: Get the filtered image feature g(y i ); Among them, W g represents the training weight matrix parameter, P represents the output feature layer index, ξ1 and ξ2 represent bilateral filter constants, and x c is the total perturbation value of the image in the dataset, is the estimated value of the total perturbation of the image in the dataset.

5. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The method of re-extracting features from the image feature model using a normalization function includes: Based on the filtered image features g(y i ), using non-local mean and bilateral filtering methods to filter out information with overlapping and chaotic features in the image features; And use the normalization function to simplify the filtered image features to obtain a new image feature model in, is the total perturbation estimate of the image in the dataset, is the local operator pose spatial feature estimation value, and β is the feature map of tomato sampled leaves at different scales.

6. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The method of calculating the observation error using a model compensation observer for the image feature model after feature re-extraction includes: Get the actual value ω of the local operator pose space feature in the matrix expression model c and local operator pose space feature estimates Calculate ω separately c and The distance between the two and the diagonal distance, and ω c and The distance between them is recorded as D, and the diagonal distance is recorded as Get a new image feature model Computing Observation Errors Using Model Compensated Observers Among them, α represents the enhanced feature layer of the tomato sampled leaves, v represents filtering smoothing, N represents the correction factor, and sig(g) represents the Gaussian radial basis function.

7. The method for rapid detection of tomato leaf diseases in real time data according to claim 1, characterized in that: The method of designing a fixed-time control input vector of a volumetric sliding mode filter according to the observation error and constructing a tomato disease detection model in combination with a Lyapunov function includes: Extract leaf image features that classify tomato leaves into green ripe stage, color change stage and red ripe stage in the data set respectively, confirm that the output feature layer index of the green ripe stage leaf image is w, the output feature layer index of the color change stage leaf image is ρ, and the output feature layer index of the red ripe stage leaf image is h; Get the observation error The gamma function is used to obtain the fixed-time control input vector χ of the cubature sliding mode filter: Where D1 and D2 represent the Euclidean distance between the actual value of the local operator pose space feature and the estimated value of the local operator pose space feature, respectively; The tomato disease detection model V(χ) is generated using the Lyapunov function: Among them, μ(χ) represents the selection accuracy parameter, θ(χ) represents the recall function, represents the average precision mean measurement function, g(y i ) represents the filtered image features, and i and n both represent integers.

Images