Tomato growth analysis method and system based on image recognition

Abstract

The present application relates to the technical field of image recognition, and particularly relates to a tomato growth analysis method and system based on image recognition, which comprises a tomato image acquisition module, an image data processing module, a model establishing module, a data analysis processing module, a model deployment module, an environment data acquisition module and an identification optimization module; the tomato image acquisition module is used for collecting image information of different growth stages of tomato plants, the image information represents tomato plant morphology and leaf image data, and the image information is sent to the image data processing module; the environment data acquisition module is used for collecting environment data when the tomato image is collected, the environment data is the light intensity, weather and season information at the moment when the tomato image is collected, and the environment data is sent to the image data processing module; meanwhile, the tomato is quickly identified and qualified, so as to facilitate the identification of the disease and pest area of the leaf, and reduce the interference of the image recognition caused by the difference between the tomato plants.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a tomato growth analysis method and system based on image recognition. Background Technology

[0002] Tomatoes are highly nutritious and edible, and are one of my country's important economic crops. With the increasing demand for food, the planting area and yield of tomatoes are constantly increasing. However, tomatoes are a crop that is susceptible to disease. Disease infection can easily lead to large-scale damage to the plant and cause irreparable economic losses. Most tomato diseases start from the leaves and spread to the entire plant. Early warning of tomato leaf diseases can prevent the spread of diseases and reduce economic losses.

[0003] However, traditional feature fusion methods use fixed weighting operations to enhance local features and suppress the expression of interfering features. But the differences in disease category images affect the model's generalization ability, resulting in low classification efficiency and accuracy. Image classification technology is widely used in the agricultural field, especially in disease detection and classification, and is more efficient and accurate than traditional manual methods.

[0004] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a tomato growth analysis method and system based on image recognition to address the aforementioned technical deficiencies. This invention acquires a large number of tomato plant images and environmental data, preprocesses the data, and models tomato growth analysis and disease identification separately. By cascading the acquired tomato plant disease identification model and tomato plant morphology identification model, the output of the tomato plant morphology identification model is passed to the tomato plant disease identification model as input, and a quasi-model is generated and sent to the identification optimization module for model cascading iteration. This allows the model to quickly identify the location of tomato diseases and the corresponding pest and disease categories for labeling, while also better distinguishing interference backgrounds in tomato images.

[0006] The objective of this invention can be achieved through the following technical solution: a tomato growth analysis system based on image recognition, comprising a tomato image acquisition module, an image data processing module, a model building module, a data analysis and processing module, a model deployment module, an environmental data acquisition module, and a recognition optimization module;

[0007] The tomato image acquisition module is used to collect image information of tomato plants at different growth stages. The image information represents the morphology of the tomato plant and leaf image data, and the image information is sent to the image data processing module.

[0008] The environmental data acquisition module is used to collect environmental data during tomato image acquisition. The environmental data includes the time of tomato image acquisition, external light intensity, weather and seasonal information, and sends the environmental data to the image data processing module.

[0009] After receiving information from the tomato image acquisition module and the environmental data acquisition module, the image data processing module marks the location of tomato diseases and the types of pests and diseases in the image data, and sends them to the data analysis and processing module and the model building module.

[0010] The model building module, after acquiring information from the image data processing module, uses the tomato growth analysis unit to acquire tomato growth environment data and tomato plant morphology, and combines it with training sets of external light intensity, weather, and seasonal information. Based on the ARIMA model, a second training set of light intensity, weather, and seasonal information is added to obtain a tomato plant morphology recognition model. The disease recognition unit acquires leaf image dataset one, and based on the Faster R-CNN model, it adds detection of whether tomato plants are affected by diseases and the corresponding pest and disease categories to obtain a tomato plant disease recognition model.

[0011] After receiving information from the image data processing module and data from the model building module, the data analysis and processing module performs model cascading on the acquired tomato plant disease identification model and tomato plant morphology identification model. The model cascading is achieved by arranging the tomato plant morphology identification model and the tomato plant disease identification model in sequence. The output of the tomato plant morphology identification model is passed to the tomato plant disease identification model as input, and a quasi-model is generated and sent to the identification optimization module.

[0012] After receiving information from the data analysis and processing module, the identification and optimization module adjusts the parameter configuration of the quasi-model, rewrites the parallel processing task into parallel or distributed processing, and passes the quasi-model to the data analysis and processing module for iteration. After the quasi-model meets the required requirements, it outputs the quasi-model to the model deployment module.

[0013] After receiving the data from the recognition and optimization module, the model deployment module deploys the quasi-model, automatically processes the image and performs growth analysis, and generates corresponding reports.

[0014] Preferably, the image processing procedure of the image data processing module is as follows:

[0015] By acquiring data from the tomato image acquisition module, the data is cleaned to remove noise from the image, and the image enhancement is performed by adjusting the contrast and correcting the color. The image is divided into the background, the tomato plant and the tomato plant part, and the disease location area is extracted. The disease location and the corresponding pest and disease category are labeled, and the data enhancement is performed on the leaf samples of tomato mosaic disease category, and tomato leaf pest and disease training set one is generated.

[0016] The environmental data acquisition module receives external environmental data at the time of tomato image acquisition, including external light intensity, weather and seasonal information, and combines it with tomato plant morphology images to generate environmental data training set two.

[0017] Preferably, the process of establishing the disease identification unit in the model building module is as follows:

[0018] Faster R-CNN uses a regression network to adjust the position and size of the detection boxes, with the initial anchor box coordinates being (x...). a ,y a ,w a ,h a The predicted offset is (t) x ,t y ,t w ,t h ), w a h is the width of the anchor frame. a x is the height of the anchor frame. a The center coordinates of the anchor frame are the x-axis values ​​and the y-axis values. a t represents the y-axis value of the center coordinate of the anchor frame. x Predict the x-axis value for the center coordinate of the anchor frame, t y Predict the y-axis value for the center coordinate of the anchor frame, t w t is the predicted width of the anchor frame. h Let be the predicted height of the anchor box, and x be the x-coordinate of the center point of the actual bounding box. It refers to t w The exponential function is used as a scaling factor to transform the predicted width into the actual width. It refers to t h The exponential function is used to transform the predicted height into a scaling factor that converts it to the actual height. The actual predicted bounding box coordinates are obtained through the following formula:

[0019]

[0020] A fixed-size window is moved across the image using a sliding window technique, and a model is applied to each window for detection. The window size for a tomato preprocessed image is W×H, where W is the window width and H is the window height. The sliding window moves across the entire image, and the result is obtained using the following formula:

[0021] W_{i,j}=I[i:i+W,j:j+H]

[0022] Where W_{i,j} represents a sub-window extracted from image I, where i and j are loop variables or index variables. In each sub-window, feature extraction and classifiers are applied to detect the presence and state of tomatoes.

[0023] Then, through convolution, a weighted summation operation is performed on a local region of the image, which is derived by the following formula:

[0024]

[0025] Where I is the input image, K is the convolution kernel, (x,y) is the position of the output feature map, m and n are the radii of the convolution kernel K in the horizontal and vertical directions, respectively, and K(i,j) represents the weight value of the convolution kernel K at position (i,j). The shape of the leaves and the size of the leaf patches in the tomato image are extracted.

[0026] Preferably, the process of the tomato growth analysis unit is as follows:

[0027] The autocorrelation function (ACF) and partial autocorrelation function (PACF) represent the correlation between a time series and its lagged version, respectively. ACF indicates...

[0028]

[0029] Where, γ k The time series has a lag, k is the autocovariance, and γ0 is the variance of the time series. Through differencing operations in the ARIMA model, the non-stationary time series is transformed into a stationary one. The first difference is expressed as...

[0030]

[0031] Among them, X t This represents the observation value of the time series at time t. In a first-order difference operation, the observed value X at the current time t is represented. t Subtracting the observation from the previous time step t-1, and modeling the tomato growth analysis, seasonal differencing and seasonal AR and MA orders are used. The seasonal part of the seasonal ARIMA model is represented as follows:

[0032]

[0033] Among them, (1-φ1B-…-φ P B P In the diagram, B is a lag operator, representing the lag of the time series (φ1,…,φ). P ) is the seasonal autoregressive coefficient, representing the seasonal lag effect of the time series, (1-Φ1B s -…-Φ Q B sQ In ), B s The lag operator representing the seasonal cycle s, (Φ1,...,Φ Q Y is the seasonal autoregressive coefficient, used to capture long-term memory in seasonal effects. t This represents the seasonally adjusted time series value at time t, (1+θ1B+...+θ) q B q In ), (θ1,...,θ q (1+Θ1B) is the seasonal moving average coefficient, used to handle seasonal random fluctuations. s +...+Θ Q B sQ In ), Θ1,...,Θ Q It is a seasonal moving average coefficient used to handle long-term seasonal dependence. This represents the white noise error term at time t, which is the random fluctuation component not explained by the model.

[0034] Preferably, the analysis process of the data analysis and processing module is as follows:

[0035] Cascaded convolutional and pooling layers are used to extract features from an image, progressively reducing the spatial size of the feature map. An image of a tomato plant morphology is represented as X1, where each element represents a pixel value. A convolutional kernel K1 is used to perform a convolution operation on the image to obtain the feature map F1.

[0036]

[0037] Where (i,j) is the pixel position in the feature map, (m,n) is the index of the convolution kernel, X1(i+m,j+n) is the pixel value in the image, and K1(m,n) is the weight of the convolution kernel, which is used to reduce the spatial size of the feature map while retaining the main features. The feature map F1 is obtained after convolution.

[0038] Preferably, the optimized recognition process in the recognition optimization module is as follows:

[0039] By selecting tomato growth characteristics and pest and disease characteristics, new original disease identification data and tomato growth data features are constructed from the original disease identification data and tomato growth data. Combined with the prediction results of the model building module, the prediction results of different feature sets are fused through the model. Training data is increased by rotating, flipping, and scaling, and noise is added or new feature samples are generated to enrich the diversity of the data.

[0040] The beneficial effects of this invention are as follows:

[0041] (1) This invention utilizes tomato images and environmental data at the time of image acquisition, and adds light intensity, weather and seasonal information to the ARI MA model to model the growth analysis of tomatoes, so that the established model is more in line with the four-season growth pattern of tomato plants, and the model has better periodic perception. At the same time, it can quickly identify and characterize tomatoes, which is conducive to the identification of disease and pest areas on leaves, and reduces the interference of image recognition caused by differences between tomato plants.

[0042] (2) The present invention cascades the obtained tomato plant disease identification model and tomato plant morphology identification model, and transmits the output of the tomato plant morphology identification model to the tomato plant disease identification model as input, thereby further reducing the overall identification range of the model, reducing the model identification time, and improving the model identification speed. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings;

[0044] Figure 1 This is a reference diagram of the method of the present invention;

[0045] Figure 2 This is a flowchart of the system of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Example 1:

[0048] Please see Figures 1 to 2 As shown, this invention is a tomato growth analysis method based on image recognition, comprising the following steps:

[0049] Step 1: Acquire tomato image data, archive and save the environmental data at the time of acquisition, transfer the data to Step 2, and match the tomato image data with the environmental data at the time of acquisition.

[0050] Step 2: Acquire the archived data, preprocess the data, summarize and categorize the environmental information corresponding to the images, and transfer the data to Steps 3 and 4 to perform data augmentation on the few tomato mosaic disease leaf samples.

[0051] Step 3: By acquiring preprocessed data, model training is performed to establish a tomato growth analysis model and a disease identification model, respectively. The data is then transferred to Step 4, and the disease location areas are extracted, and the disease locations and their respective pest and disease categories are labeled.

[0052] Step 4: Input the plant morphology labels and morphology images output by the tomato growth analysis model into the disease identification model through cascading multilayer layers to reduce the data processing time and scope, and then transfer the data to Step 5, where cascading multilayer convolutional and pooling layers are used to extract features from the image and gradually reduce the spatial size of the feature map.

[0053] Step 5: By selecting and adjusting the model parameters, and combining the prediction results of the model building module, the prediction results of different feature sets are fused through the model. The original disease identification data and tomato growth data are used to construct new original disease identification data and tomato growth data features. The prediction results of the model building module are then combined with the prediction results of different feature sets through the model.

[0054] A tomato growth analysis system based on image recognition includes a tomato image acquisition module, an image data processing module, a model building module, a data analysis and processing module, a model deployment module, an environmental data acquisition module, and a recognition optimization module.

[0055] The tomato image acquisition module collects image information of tomato plants at different growth stages, representing tomato plant morphology and leaf image data, and sends the image information to the image data processing module. The environmental data acquisition module collects environmental data during tomato image acquisition, including the time of image acquisition, external light intensity, weather, and seasonal information, and sends the environmental data to the image data processing module. After receiving information from the tomato image acquisition module and the environmental data acquisition module, the image data processing module labels the location of tomato diseases and the corresponding pest / disease categories in the image data, and sends the data to the data analysis and processing module and the model building module.

[0056] The model building module, after acquiring information from the image data processing module, uses the tomato growth analysis unit to acquire tomato growth environment data and tomato plant morphology, and combines it with training sets of external light intensity, weather, and seasonal information. Based on the ARIMA model, a second training set of light intensity, weather, and seasonal information is added to obtain a tomato plant morphology recognition model. The disease recognition unit acquires leaf image dataset one, and based on the Faster R-CNN model, it adds detection of whether tomato plants are affected by diseases and the corresponding pest and disease categories to obtain a tomato plant disease recognition model.

[0057] After receiving information from the image data processing module and data from the model building module, the data analysis and processing module performs model cascading on the acquired tomato plant disease identification model and tomato plant morphology identification model. The model cascading is achieved by arranging the tomato plant morphology identification model and the tomato plant disease identification model in sequence. The output of the tomato plant morphology identification model is passed to the tomato plant disease identification model as input, and a quasi-model is generated and sent to the identification optimization module.

[0058] After receiving information from the data analysis and processing module, the recognition optimization module adjusts the parameter configuration of the quasi-model, rewrites the parallel processing task into a parallel or distributed manner, and passes the quasi-model to the data analysis and processing module for iteration. Once the quasi-model meets the required requirements, it is output to the model deployment module. The model deployment module, after receiving data from the recognition optimization module, deploys the quasi-model, automatically processes the image and performs growth analysis, and generates corresponding reports. The image processing process of the image data processing module is as follows:

[0059] By acquiring data from the tomato image acquisition module, the data is cleaned to remove noise from the images, and the image enhancement is performed by adjusting the contrast and correcting the color. The images are divided into background, tomato plant and tomato plant parts, and the disease location area is extracted. The disease location and the corresponding pest and disease category are labeled, and the data enhancement is performed on the leaf samples of tomato mosaic disease category, and tomato leaf pest and disease training set 1 is generated.

[0060] The environmental data acquisition module receives external environmental data at the time of tomato image acquisition, including external light intensity, weather, and seasonal information, and combines this data with tomato plant morphology images to generate environmental data training set two. The tomato growth analysis unit's process is as follows: the autocorrelation function (ACF) and partial autocorrelation function (PACF) represent the correlation between the time series and its lagged version, respectively. ACF represents...

[0061]

[0062] Where, γ kThe time series has a lag, k is the autocovariance, and γ0 is the variance of the time series. Through differencing operations in the ARIMA model, the non-stationary time series is transformed into a stationary one. The first difference is expressed as...

[0063]

[0064] Among them, X t This represents the observation value of the time series at time t. In a first-order difference operation, the observed value X at the current time t is represented. t Subtracting the observation from the previous time step t-1, and modeling the tomato growth analysis, seasonal differencing and seasonal AR and MA orders are used. The seasonal part of the seasonal ARIMA model is represented as follows:

[0065]

[0066] Among them, (1-φ1B-…-φ P B P In the diagram, B is a lag operator, representing the lag of the time series (φ1,…,φ). P ) is the seasonal autoregressive coefficient, representing the seasonal lag effect of the time series, (1-Φ1B s -…-Φ Q B sQ In ), B s The lag operator representing the seasonal cycle s, (Φ1,…,Φ Q Y is the seasonal autoregressive coefficient, used to capture long-term memory in seasonal effects. t This represents the seasonally adjusted time series value at time t, (1+θ1B+…+θ q B q In ), (θ1,…,θ q (1+Θ1B) is the seasonal moving average coefficient, used to handle seasonal random fluctuations. s +…+Θ Q B sQ In ), Θ1,…,Θ Q It is a seasonal moving average coefficient used to handle long-term seasonal dependence. t This represents the white noise error term at time t, i.e., the random fluctuations that are not explained by the model.

[0067] Image information of tomato plants at different growth stages is collected, representing tomato plant morphology and leaf image data. This image information is sent to the image data processing module. Environmental data at the time of tomato image acquisition, including the moment of image acquisition, external light intensity, weather, and seasonal information, is also collected and sent to the image data processing module. After receiving information from the tomato image acquisition module and the environmental data acquisition module, the location of tomato diseases and their corresponding pest and disease categories in the image data are labeled and sent to the data analysis and processing module and the model building module. The model building module, after acquiring information from the image data processing module, uses the tomato growth analysis unit to acquire tomato growth environment data and tomato plant morphology. It combines this data with a training set of external light intensity, weather, and seasonal information, and adds a second training set of light intensity, weather, and seasonal information to the ARIMA model to obtain a tomato plant morphology recognition model.

[0068] Example 2:

[0069] The process of establishing the disease identification unit in the model building module is as follows:

[0070] Faster R-CNN uses a regression network to adjust the position and size of the detection boxes, with the initial anchor box coordinates being (x...). a ,y a ,w a ,h a The predicted offset is (t) x ,t y ,t w ,t h ), w a h is the width of the anchor frame. a x is the height of the anchor frame. a The center coordinates of the anchor frame are the x-axis values ​​and the y-axis values. a t represents the y-axis value of the center coordinate of the anchor frame. x Predict the x-axis value for the center coordinate of the anchor frame, t y Predict the y-axis value for the center coordinate of the anchor frame, t w t is the predicted width of the anchor frame. h Let be the predicted height of the anchor box, and x be the x-coordinate of the center point of the actual bounding box. It refers to t w The exponential function is used as a scaling factor to transform the predicted width into the actual width. It refers to t h The exponential function is used to transform the predicted height into a scaling factor that converts it to the actual height. The actual predicted bounding box coordinates are obtained through the following formula:

[0071]

[0072] A fixed-size window is moved across the image using a sliding window technique, and a model is applied to each window for detection. The window size for a tomato preprocessed image is W×H, where W is the window width and H is the window height. The sliding window moves across the entire image, and the result is obtained using the following formula:

[0073] W_{i,j}=I[i:i+W,j:j+H]

[0074] Where W_{i,j} represents a sub-window extracted from image I, where i and j are loop variables or index variables. In each sub-window, feature extraction and classifiers are applied to detect the presence and state of tomatoes. A weighted summation operation is performed on local regions of the image through convolution, derived by the following formula:

[0075]

[0076] Where I is the input image, K is the convolution kernel, (x,y) is the position of the output feature map, m and n are the radii of the convolution kernel K in the horizontal and vertical directions, respectively, and K(i,j) represents the weight value of the convolution kernel K at position (i,j). The data is used to extract the shape of the leaves and the size of the leaf patches in the tomato image. The analysis process of the data analysis module is as follows: cascaded convolutional layers and pooling layers are used to extract features from the image, and the spatial size of the feature map is gradually reduced. An image of a tomato plant morphology is represented as X1, where each element represents a pixel value. Convolution kernel K1 is used to perform a convolution operation on the image to obtain the feature map F1.

[0077]

[0078] Where (i,j) is the pixel position in the feature map, (m,n) is the index of the convolution kernel, X1(i+m,j+n) is the pixel value in the image, and K1(m,n) is the weight of the convolution kernel, used to reduce the spatial size of the feature map while retaining the main features. The feature map F1 obtained after convolution is then subjected to max pooling.

[0079] P(i,j)=max m,n∈pooling region F(i·s+m,j·s+n)

[0080] Where P is the pooled feature map and s is the stride of the pooling window;

[0081] The optimized recognition process in the recognition optimization module is as follows:

[0082] By selecting tomato growth characteristics and pest and disease characteristics, new original disease identification data and tomato growth data features are constructed from the original disease identification data and tomato growth data. Combined with the prediction results of the model building module, the prediction results of different feature sets are fused through the model. Training data is increased by rotating, flipping, and scaling, and noise is added or new feature samples are generated to enrich the diversity of the data.

[0083] The disease identification unit acquires leaf image datasets based on Faster. Building upon the R-CNN model, a tomato plant disease identification model is created by detecting whether tomato plants are affected by diseases and identifying the corresponding pest or disease category. This model is then cascaded with the tomato plant morphology identification model. The cascading process involves sequentially arranging the tomato plant morphology and disease identification models. As the network depth increases, multiple convolutional and pooling layers are stacked. Each convolutional layer increases the complexity and abstraction level of the features, while each pooling layer gradually reduces the size of the feature maps while maintaining the expressive power of important features. This allows the network to learn more abstract and complex features at a higher level. Simultaneously, the pooling layers reduce the spatial size of the feature maps, enabling the network to handle input images of different sizes and achieve good generalization ability. The output of the tomato plant morphology identification model is passed to the tomato plant disease identification model as input. The quasi-model is then fed into the data analysis and processing module for iteration. Once the quasi-model meets the required standards, it is output to the model deployment module for deployment. The module automatically processes images, performs growth analysis, and generates corresponding reports.

[0084] The above formulas are all derived from software simulation using a large amount of data, and are selected to be close to the actual values. The coefficients in the formulas are set by those skilled in the art according to the actual situation. The above are only preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or changes made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A tomato growth analysis system based on image recognition, characterized in that, It includes a tomato image acquisition module, an image data processing module, a model building module, a data analysis and processing module, a model deployment module, an environmental data acquisition module, and a recognition and optimization module; The tomato image acquisition module is used to collect image information of tomato plants at different growth stages. The image information represents the morphology of the tomato plant and leaf image data, and the image information is sent to the image data processing module. The environmental data acquisition module is used to collect environmental data during tomato image acquisition. The environmental data includes the time of tomato image acquisition, external light intensity, weather and seasonal information, and sends the environmental data to the image data processing module. After receiving information from the tomato image acquisition module and the environmental data acquisition module, the image data processing module marks the location of tomato diseases and the types of pests and diseases in the image data, and sends them to the data analysis and processing module and the model building module. The model building module, after acquiring information from the image data processing module, uses the tomato growth analysis unit to acquire tomato growth environment data and tomato plant morphology, and combines it with training sets of external light intensity, weather, and seasonal information. Based on the ARIMA model, a second training set of light intensity, weather, and seasonal information is added to obtain a tomato plant morphology recognition model. The disease recognition unit acquires leaf image dataset one, and based on the Faster R-CNN model, it adds detection of whether tomato plants are affected by diseases and the corresponding pest and disease categories to obtain a tomato plant disease recognition model. After receiving information from the image data processing module and data from the model building module, the data analysis and processing module performs model cascading on the acquired tomato plant disease identification model and tomato plant morphology identification model. The model cascading is achieved by arranging the tomato plant morphology identification model and the tomato plant disease identification model in sequence. The output of the tomato plant morphology identification model is passed to the tomato plant disease identification model as input, and a quasi-model is generated and sent to the identification optimization module. After receiving information from the data analysis and processing module, the identification and optimization module adjusts the parameter configuration of the quasi-model, rewrites the parallel processing task into parallel or distributed processing, and passes the quasi-model to the data analysis and processing module for iteration. After the quasi-model meets the required requirements, it outputs the quasi-model to the model deployment module. After receiving the data from the recognition and optimization module, the model deployment module deploys the quasi-model, automatically processes the image and performs growth analysis, and generates corresponding reports.

2. The tomato growth analysis system based on image recognition according to claim 1, characterized in that, The image processing procedure of the image data processing module is as follows: By acquiring data from the tomato image acquisition module, the data is cleaned to remove noise from the image, and the image enhancement is performed by adjusting the contrast and correcting the color. The image is divided into the background, the tomato plant and the tomato plant part, and the disease location area is extracted. The disease location and the corresponding pest and disease category are labeled, and the data enhancement is performed on the leaf samples of tomato mosaic disease category, and tomato leaf pest and disease training set one is generated. The environmental data acquisition module receives external environmental data at the time of tomato image acquisition, including external light intensity, weather and seasonal information, and combines it with tomato plant morphology images to generate environmental data training set two.

3. The tomato growth analysis system based on image recognition according to claim 1, characterized in that, The process of establishing the disease identification unit in the model building module is as follows: Faster R-CNN uses a regression network to adjust the position and size of the detection boxes, with the initial anchor box coordinates being (x...). a ,y a ,w a ,h a The predicted offset is (t) x ,t y ,t w ,t h ), w a h is the width of the anchor frame. a x is the height of the anchor frame. a The center coordinates of the anchor frame are the x-axis values ​​and the y-axis values. a t represents the y-axis value of the center coordinate of the anchor frame. x Predict the x-axis value for the center coordinate of the anchor frame, t y Predict the y-axis value for the center coordinate of the anchor frame, t w t is the predicted width of the anchor frame. h Let be the predicted height of the anchor box, and x be the x-coordinate of the center point of the actual bounding box. It refers to t w The exponential function is used as a scaling factor to transform the predicted width into the actual width. It refers to t h The exponential function is used to transform the predicted height into a scaling factor of the actual height. The actual predicted bounding box coordinates are obtained using the following formula: A fixed-size window is moved across the image using a sliding window technique, and a model is applied to each window for detection. The window size for a tomato preprocessed image is W×H, where W is the window width and H is the window height. The sliding window moves across the entire image, and the result is obtained using the following formula: W_{i,j}=I[i:i+W,j:j+H] Where W_{i,j} represents a sub-window extracted from image I, where i and j are loop variables or index variables. In each sub-window, feature extraction and classifiers are applied to detect the presence and state of tomatoes. Then, through convolution, a weighted summation operation is performed on a local region of the image, which is derived by the following formula: Where I is the input image, K is the convolution kernel, (x,y) is the position of the output feature map, m and n are the radii of the convolution kernel K in the horizontal and vertical directions, respectively, and K(i,j) represents the weight value of the convolution kernel K at position (i,j). The shape of the leaves and the size of the leaf patches in the tomato image are extracted.

4. The tomato growth analysis system based on image recognition according to claim 1, characterized in that, The process of the tomato growth analysis unit is as follows: The autocorrelation function (ACF) and partial autocorrelation function (PACF) represent the correlation between a time series and its lagged version, respectively. The ACF is expressed as: Where, γ k The time series is lagged, k is the autocovariance, and γ0 is the variance of the time series. Through differencing operations in the ARIMA model, the non-stationary time series is transformed into a stationary time series. The first-order differencing is expressed as: ▽X t =X t -X t-1 Among them, X t X represents the observation value of the time series at time t. t In a first-order difference operation, the observed value X at the current time t is represented. t Subtracting the observation value from the previous time step t-1, and modeling the tomato growth analysis, seasonal differencing and seasonal AR and MA orders are used for processing. The seasonal part of the seasonal ARIMA model is expressed as follows: (1-φ1B-...-φ P B P )(1-Φ1B s -...-Φ Q B sQ )Y t = (1+θ1B+...+θ) q B q )(1+Θ1B s +...+Θ Q B sQ )ò t Among them, (1-φ1B-...-φ P B P In the diagram, B is a lag operator, representing the lag in the time series (φ1,...,φ). P ) is the seasonal autoregressive coefficient, representing the seasonal lag effect of the time series, (1-Φ1B s -...-Φ Q B sQ In ), B s The lag operator representing the seasonal cycle s, (Φ1,...,Φ Q Y is the seasonal autoregressive coefficient, used to capture long-term memory in seasonal effects. t This represents the seasonally adjusted time series value at time t, (1+θ1B+...+θ) q B q In ), (θ1,...,θ q (1+Θ1B) is the seasonal moving average coefficient, used to handle seasonal random fluctuations. s +...+Θ Q B sQ In ), Θ1,...,Θ Q It is a seasonal moving average coefficient used to handle long-term seasonal dependence. t This represents the white noise error term at time t, which is the random fluctuation component not explained by the model.

5. The tomato growth analysis system based on image recognition according to claim 1, characterized in that, The analysis process of the data analysis and processing module is as follows: Cascaded convolutional and pooling layers are used to extract features from an image, progressively reducing the spatial size of the feature map. An image of a tomato plant morphology is represented as X1, where each element represents a pixel value. A convolutional kernel K1 is used to convolve the image, resulting in feature map F1. Where (i,j) is the pixel position in the feature map, (m,n) is the index of the convolution kernel, X1(i+m,j+n) is the pixel value in the image, and K1(m,n) is the weight of the convolution kernel, which is used to reduce the spatial size of the feature map while retaining the main features. The feature map F1 is obtained after convolution.

6. The tomato growth analysis system based on image recognition according to claim 1, characterized in that, The optimized recognition process in the recognition optimization module is as follows: By selecting tomato growth characteristics and pest and disease characteristics, new original disease identification data and tomato growth data features are constructed from the original disease identification data and tomato growth data. Combined with the prediction results of the model building module, the prediction results of different feature sets are fused through the model. Training data is increased by rotating, flipping, and scaling, and noise is added or new feature samples are generated.

7. A tomato growth analysis method based on image recognition, wherein the method is applied to the tomato growth analysis system based on image recognition as described in any one of claims 1-6, characterized in that, Includes the following steps: Step 1: Acquire tomato image data, archive and save the environmental data at the time of acquisition, and transfer the data to Step 2; Step 2: Acquire the archived data, preprocess the data, summarize and categorize the environmental information corresponding to the images, and transfer the data to Step 3 and Step 4; Step 3: By acquiring preprocessed data, model training is performed to establish a tomato growth analysis model and a disease identification model, and the data is then transferred to Step 4; Step 4: Input the plant morphology labels and morphological images output by the tomato growth analysis model into the disease identification model through multi-level cascading to reduce data processing time and scope, and then transfer the data to Step 5; Step 5: By selecting and tuning the model, and combining the prediction results of the model building module, the prediction results of different feature sets are fused through the model.

Images