Citrus shoot length and leaf color state detection method and system based on shade guide correction

Abstract

The invention discloses a citrus shoot length and leaf color state detection method and system based on colorimetric plate correction, and relates to the field of crop monitoring, and the method comprises the steps: fixing a colorimetric plate below a citrus shoot, and synchronously collecting image information containing a complete shoot region and a colorimetric plate region by using a mobile phone; on the basis of grid lines on the colorimetric plate, determining the branch length and the growth stage of the citrus; performing color calibration based on a standard color block on the colorimetric plate, generating a corrected image, and constructing a training set in combination with the branch length and growth stage information; building a neural network model based on a deep learning technology, and performing model training by using the training set to obtain an evaluation model of citrus branch tip and leaf development; and performing state identification on the corrected image of the to-be-detected branch tip by using the evaluation model of the citrus branch tip and leaf development, obtaining the current branch tip length, and judging whether the current leaf color is in a reasonable range or not. According to the invention, the growth condition of the citrus branch tip can be clearly and accurately known, and the identification speed and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of crop monitoring, and more particularly to a method and system for detecting citrus branch length and leaf color status based on colorimetric plate correction. Background Art

[0002] Citrus fruits occupy a significant share of the fruit market. With increasing demand for high-quality citrus, the scale and output of citrus cultivation are expanding year by year. Citrus growth directly impacts yield and quality. Traditional citrus cultivation management relies primarily on manual observation and empirical judgment. This approach is not only inefficient but also susceptible to subjective factors, making precise management difficult.

[0003] Branch length reflects the growth stage and nutrient distribution capacity of citrus trees and serves as a basis for pruning decisions. Leaf color, on the other hand, reflects photosynthetic efficiency and physiological abnormalities and serves as an important reference for yield prediction. Therefore, rapid and accurate detection of citrus branch length and leaf color is crucial for the precise management of citrus cultivation. Currently, numerous methods exist for detecting these characteristics. Analyzing citrus images using computer vision and image processing techniques is a current research hotspot and development trend, but many challenges remain in practical applications.

[0004] When it comes to measuring branch length, image recognition-based measurement methods often suffer from problems such as excessive blurring and blurred subjects due to complex field backgrounds, making it difficult to accurately identify branch boundaries and severely impacting measurement accuracy. Furthermore, due to the lack of unified standards, measurement results from different image measurement systems lack comparability. When it comes to detecting leaf color, environmental factors such as lighting and weather conditions significantly influence color. Citrus leaf color varies significantly under varying light intensities, angles, and colors, creating significant challenges for image composition and exposure, severely impacting color detection accuracy.

[0005] Therefore, how to prevent the quality of photos from being affected by field lighting and photographic equipment, understand the growth of citrus more clearly and accurately, and improve recognition speed and accuracy are technical problems that technical personnel in this field urgently need to solve. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for detecting the length of citrus branches and leaves and the color status of leaves based on colorimetric plate correction, which solves the problems existing in the background technology.

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

[0008] A method for detecting citrus branch length and leaf color status based on colorimetric plate correction comprises the following steps:

[0009] Fix a colorimetric plate below the citrus branches, and use a mobile phone to simultaneously capture image information including the complete branch area and the colorimetric plate area;

[0010] Based on the grid lines on the color chart, confirm the length of the branches and the growth stage of the citrus;

[0011] Preprocess the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image;

[0012] Construct a training set based on the corrected image, branch length, and growth stage information;

[0013] A neural network model was built based on deep learning technology and trained using a training set until the network converged, resulting in an evaluation model for citrus shoot and leaf development.

[0014] The evaluation model of citrus branch and leaf development is used to perform state recognition on the corrected image of the branch to be tested, obtain the current branch length, and determine whether the current leaf color is within a reasonable range.

[0015] Optionally, fix a colorimetric plate below the citrus branches, specifically:

[0016] Use a fixed bracket with an adjustable buckle, clamp one end of the fixed bracket to the citrus tree trunk with the buckle, install a horizontally placed flat plate on the other end, and magnetically fix the colorimetric plate to the flat plate so that the colorimetric plate is directly below the citrus branch tip and remains level with the ground.

[0017] Optionally, the colorimetric plate is a rectangular flat plate structure, and the surface is divided into two areas; one area is an area with equally spaced grid lines, and the grid lines are squares with black lines, which are used to assist in confirming the length of branches; the other area is a standard color block area, and the color of the color block covers the color range of citrus leaves in different health conditions, which is used to perform color calibration on the collected images.

[0018] Optionally, determining the length of the shoots and the growth stage of the citrus fruits may include the following steps:

[0019] The canny edge detection algorithm is used to identify the edges of the colorimetric plate grid lines and the outline edges of the citrus branches.

[0020] The mutually perpendicular grid lines on the colorimetric plate are mapped to the parameter space through Hough transform to construct the grid line coordinate system;

[0021] According to the position of the citrus branch in the grid coordinate system, the pixel length of the citrus branch in the image is calculated;

[0022] Using the ratio between the side length of the grid lines on the color chart and the image pixels, the pixel length is converted into the actual physical length to obtain the branch length.

[0023] The length of the branches and shoots was compared with the length range of the corresponding varieties in the pre-established citrus growth database to determine the growth stage of the citrus.

[0024] Optionally, preprocess the image information, specifically:

[0025] Gaussian filtering is used to process the image to retain the main edge and structural information of the image;

[0026] The image is processed by histogram equalization method to highlight the image details;

[0027] The region of interest (ROI) in the image, including the citrus branches and the colorimetric plate, was determined and cropped, and the cropped image was scaled to an appropriate size using bilinear interpolation.

[0028] Optionally, color calibration is performed based on a standard color block on a colorimetric plate, specifically including the following steps:

[0029] Segment the standard color block area of ​​the colorimetric plate from the preprocessed image, and extract the standard color block image using the image segmentation algorithm;

[0030] Obtain the actual color value of each standard color block and compare it with the known standard color value, and calculate the color deviation using the color difference formula;

[0031] The color value of each pixel in the image is corrected based on the color deviation, and the color is calculated and updated pixel by pixel to generate a corrected image.

[0032] Optionally, the specific method of constructing the training set is:

[0033] Convert the corrected images into RGB format and assign a unique identifier to each corrected image using the image file name;

[0034] Record the shoot length in numerical form, convert the growth stage information into category labels, and establish a correspondence with the corresponding image identifiers;

[0035] Create an image folder to store the corrected images, and record the correspondence between image identifiers, branch length values, and growth stage labels in CSV format.

[0036] Optionally, the neural network model adopts a multi-task parallel network architecture, including: a feature extraction network, a branch length prediction branch, and a leaf color state classification branch;

[0037] The feature extraction network uses the ResNet18 network to extract features from the input rectified image;

[0038] The branch length prediction branch is used to extract spatial features related to branch length through the convolution layer and map the spatial features into predicted values ​​of branch length through the regression layer;

[0039] The leaf color state classification branch is used to extract features related to the leaf color state through convolutional layers and pooling layers, and use the softmax activation function to output the probability of each leaf color state category.

[0040] Optionally, the specific steps for model training are:

[0041] Set the learning rate, batch size, and number of training rounds, input the training set into the neural network model in batches, and obtain the predicted branch length value and leaf color state predicted probability distribution;

[0042] Calculate the mean square error loss of branch length prediction and the cross entropy loss of leaf color state classification respectively, and sum them according to the corresponding weights to get the total loss;

[0043] Calculate the gradient of the total loss with respect to the model parameters, and use the optimizer to update the model parameters according to the gradient;

[0044] After completing all training rounds, the model parameters with the best performance were saved to obtain the evaluation model of citrus shoot and leaf development.

[0045] A citrus branch length and leaf color state detection system based on colorimetric plate correction, which executes any of the above-mentioned citrus branch length and leaf color state detection methods based on colorimetric plate correction, comprising:

[0046] The image acquisition module is used to fix the colorimetric plate below the citrus branches and use the mobile phone to synchronously capture image information including the complete branch area and the colorimetric plate area;

[0047] The data acquisition module is used to confirm the length of branches and the growth stage of citrus fruits through the grid lines on the colorimetric plate;

[0048] An image processing module is used to pre-process the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image;

[0049] The set construction module is used to construct the training set based on the corrected image, branch length and growth stage information;

[0050] The model building and training module is used to build a neural network model using deep learning technology and train the model using a training set until the network converges to obtain an evaluation model for citrus shoot and leaf development;

[0051] The real-time detection module is used to use the evaluation model of citrus branch and leaf development to perform state recognition on the corrected image of the branch to be tested, obtain the current branch length and determine whether the current leaf color is within a reasonable range.

[0052] It can be seen from the above technical solutions that compared with the existing technology, the present invention provides a method and system for detecting the length of citrus branches and leaves color status based on colorimetric plate correction. The color of the collected images is calibrated using a colorimetric plate, which can prevent the quality of the photos from being affected by field lighting and photographic equipment. The multi-task parallel network architecture is adopted to improve the learning efficiency of the model, and the citrus growth status detection results are more accurate and reliable, providing a universal monitoring solution for citrus cultivation. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0054] Figure 1 A flow chart of the method for detecting citrus branch length and leaf color status based on colorimetric plate correction provided by the present invention;

[0055] Figure 2 This is a structural diagram of the citrus branch length and leaf color status detection system based on colorimetric plate correction provided by the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0057] In agricultural production, accurately understanding the length of citrus branches and the color of leaves is crucial for evaluating citrus growth conditions, determining growth stages, and monitoring pests and diseases. In order to prevent the quality of photos from being affected by field lighting and photographic equipment, the present invention discloses a method for detecting the length of citrus branches and the color of leaves based on colorimetric plate correction. Figure 1 As shown, the following steps are included:

[0058] A colorimetric plate is fixed below the citrus branches, and a mobile phone is used to simultaneously capture image information covering the entire branch area and the colorimetric plate area, ensuring that the captured image covers the key areas to be inspected. In this embodiment, the use of a mobile phone for shooting is simple and easy to operate, without the need for professional image acquisition equipment, thus reducing monitoring costs, improving the efficiency and flexibility of data acquisition, and facilitating promotion and application.

[0059] Based on the grid lines on the color chart, confirm the length of the branches and the growth stage of the citrus;

[0060] Preprocess the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image;

[0061] Construct a training set based on the corrected image, branch length, and growth stage information;

[0062] A neural network model was built based on deep learning technology and trained using a training set until the network converged, resulting in an evaluation model for citrus shoot and leaf development.

[0063] Using a model for evaluating citrus branch and leaf development, the calibrated image of the branch to be tested is used to identify its state, determine the current branch length, and determine whether the current leaf color is within a reasonable range. In this embodiment, based on the monitored branch length and leaf color, precise fertilization and appropriate irrigation can be achieved, improving citrus quality and yield, promoting the digital and intelligent development of the citrus industry, and enhancing modern management.

[0064] Furthermore, a colorimetric plate is fixed below the citrus branches, specifically:

[0065] Use a fixed bracket with an adjustable clip, clamp one end of the fixed bracket to the citrus tree trunk with the clip, install a horizontally placed flat plate on the other end, and magnetically fix the colorimetric plate to the flat plate so that the colorimetric plate is directly below the citrus branch tip and remains level with the ground, ensuring that the relative position of the colorimetric plate and the branch tip is stable and convenient for image acquisition.

[0066] Furthermore, the colorimetric plate is a rectangular flat plate structure, and the surface is divided into two areas; one area is an evenly spaced grid line area, and the grid lines are squares with black lines, which are used to assist in confirming the length of branches; the other area is a standard color block area, and the color of the color block covers the color range of citrus leaves in different health conditions, which is used for color calibration of the collected images.

[0067] Furthermore, confirming the length of the branches and the growth stage of the citrus fruits specifically includes the following steps:

[0068] The Canny edge detection algorithm is used to identify the edges of the colorimetric grid lines and the outlines of citrus branches. Specifically, the image is grayscaled to reduce color information interference; the grayscale image is smoothed to reduce the impact of noise; the gradient amplitude and direction of each pixel in the image are calculated, and the maximum value suppression algorithm is used to retain the edge pixels with the largest gradient amplitude and suppress the weak response pixels of non-edge pixels; the dual threshold algorithm is used to determine strong and weak edges, thereby identifying the outlines of the colorimetric grid lines and branches, and enhancing the line and object boundary features in the image;

[0069] The mutually perpendicular grid lines on the colorimetric plate are mapped to the parameter space through Hough transform to construct a grid coordinate system. Specifically, based on the mutually perpendicular characteristics of the grid lines on the colorimetric plate, the parameters of the straight lines are searched in the horizontal and vertical directions. After the grid lines are determined, a grid coordinate system is constructed with a corner of the colorimetric plate as the coordinate origin and the mutually perpendicular grid lines as the coordinate axes.

[0070] According to the position of the citrus branch in the grid coordinate system, the pixel length of the citrus branch in the image is calculated; if the starting coordinate of the branch is (x1, y1) and the ending coordinate is (x2, y2), then the pixel length

[0071]

[0072] Using the ratio between the side length of the grid lines on the color chart and the image pixels, the pixel length is converted to the actual physical length to obtain the branch length; if the actual side length of the grid lines is L0, the pixel side length in the image is L'0, and the pixel length of the branch is L, then the actual physical length of the branch is

[0073] The branch length is compared with the length ranges of corresponding varieties in a pre-established citrus growth database to determine the growth stage of the citrus. The citrus growth database stores the branch length ranges of different citrus varieties at various growth stages. The actual branch length can be used to determine the growth stage.

[0074] Regarding the measurement of citrus branch length, this embodiment uses the grid lines of the colorimetric plate to reduce the error of manual measurement, quickly obtain a large amount of branch length data, provide a reliable basis for judging the growth stage of citrus, and help to timely grasp the growth rate of the plant and adjust management measures such as fertilization and pruning.

[0075] Furthermore, in order to solve the problems of radial distortion, tangential distortion and image distortion caused by the optical characteristics of mobile phone cameras, this embodiment uses an image processing algorithm to pre-process the image information, specifically:

[0076] Gaussian filtering is used to process the image, retaining the main edge and structural information of the image and reducing noise interference;

[0077] The image is processed by the histogram equalization method to highlight the image details, enhance the image contrast and improve the overall visual effect;

[0078] The region of interest (ROI) in the image, including the citrus branches and the colorimetric plate, was determined and cropped, and the cropped image was scaled to an appropriate size using bilinear interpolation.

[0079] Different mobile phone cameras have varying degrees of color reproduction, and the large variations in field lighting can cause the captured image colors to deviate from the actual colors. Without color correction, the model may misjudge the color of leaves due to inaccurate image colors, affecting the accurate assessment of citrus growth. Therefore, to further compensate for color deviations caused by ambient lighting and equipment differences, this embodiment performs color calibration based on standard color blocks on a colorimetric chart. The specific steps include:

[0080] Segment the standard color block area of ​​the colorimetric plate from the preprocessed image, and extract the standard color block image using the image segmentation algorithm;

[0081] Obtain the actual color value of each standard color block and compare it with the known standard color value, and calculate the color deviation using the color difference formula; specifically, the RGB color space or CIELAB color space can be used for color representation;

[0082] The color value of each pixel in the image is corrected based on the color deviation, and the color is calculated and updated pixel by pixel to generate a corrected image.

[0083] Regarding the judgment of the color status of leaves, this embodiment performs color calibration through the standard color blocks of the colorimetric plate, and combines it with a deep learning model to accurately identify whether the leaf color is normal, and then determine whether the citrus has problems such as nutritional deficiencies, pests and diseases.

[0084] Furthermore, the specific method of constructing the training set is:

[0085] Convert the corrected images into RGB format and assign a unique identifier to each corrected image using the image file name;

[0086] Record the shoot length in numerical form, convert the growth stage information into category labels, and establish a correspondence with the corresponding image identifiers;

[0087] Create an image folder to store the corrected images, and record the correspondence between image identifiers, branch length values, and growth stage labels in CSV format for easy reading and processing.

[0088] This example constructs a training set based on image data, branch length information, and growth stage labels. This allows the model to learn a rich set of features and patterns, improving the accuracy of its judgments of branch length and leaf color. Data augmentation operations such as rotation, flipping, and scaling increase data diversity, enabling the model to better adapt to different scenarios and changes, and providing strong generalization capabilities.

[0089] Furthermore, the neural network model adopts a multi-task parallel network architecture, including: feature extraction network, branch length prediction branch, and leaf color state classification branch;

[0090] The feature extraction network uses the ResNet18 network to extract features from the input rectified image. Specifically, the ResNet18 network contains multiple residual blocks, each of which consists of a convolutional layer, a batch normalization layer, and an activation function. The residual blocks introduce skip connections, allowing the network to directly learn residual mappings, which can alleviate the vanishing gradient problem in deep neural networks.

[0091] The branch length prediction branch is used to extract spatial features related to branch length through a convolutional layer and map the spatial features into predicted branch length values ​​through a regression layer. Specifically, the parameters of the convolutional layer are adjusted according to the actual situation, and a linear activation function is used in the regression layer.

[0092] The leaf color state classification branch is used to extract features related to the leaf color state through convolutional layers and pooling layers, and use the softmax activation function to output the probability of each leaf color state category.

[0093] Regarding the design of the neural network model, this embodiment adopts a multi-task parallel network architecture and a shared backbone feature extraction network. It processes the branch length prediction and leaf color status classification tasks through two branches respectively, which can improve the learning efficiency of the model, reduce the waste of computing resources, and enhance the accuracy and reliability of the model in detecting the growth status of citrus.

[0094] Furthermore, the specific steps of model training are:

[0095] Set the learning rate, batch size, and number of training rounds, input the training set into the neural network model in batches, and obtain the predicted branch length value and leaf color state predicted probability distribution;

[0096] The mean square error (MSE) loss of branch length prediction and the cross entropy loss of leaf color state classification are calculated respectively, and the total loss is obtained by weighted summation according to the corresponding weights;

[0097] Calculate the gradient of the total loss with respect to the model parameters, and use the optimizer to update the model parameters according to the gradient;

[0098] After completing all training rounds, the model parameters with the best performance were saved to obtain the evaluation model of citrus shoot and leaf development.

[0099] and Figure 1 Corresponding to the method described above, the embodiment of the present invention also provides a citrus branch length and leaf color state detection system based on colorimetric plate correction, which is used to Figure 1 The specific implementation of the method in the embodiment of the present invention provides a citrus branch length and leaf color state detection system based on colorimetric plate correction, which can be applied to computer terminals or various mobile devices, such as Figure 2 As shown, specifically including:

[0100] The image acquisition module is used to fix the colorimetric plate below the citrus branches and use the mobile phone to synchronously capture image information including the complete branch area and the colorimetric plate area;

[0101] The data acquisition module is used to confirm the length of branches and the growth stage of citrus fruits through the grid lines on the colorimetric plate;

[0102] An image processing module is used to pre-process the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image;

[0103] The set construction module is used to construct the training set based on the corrected image, branch length and growth stage information;

[0104] The model building and training module is used to build a neural network model using deep learning technology and train the model using a training set until the network converges to obtain an evaluation model for citrus shoot and leaf development;

[0105] The real-time detection module is used to use the evaluation model of citrus branch and leaf development to perform state recognition on the corrected image of the branch to be tested, obtain the current branch length and determine whether the current leaf color is within a reasonable range.

[0106] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0107] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting citrus branch length and leaf color based on colorimetric plate correction, characterized in that: The following steps are involved: Fix a colorimetric plate below the citrus branches, and use a mobile phone to simultaneously capture image information including the complete branch area and the colorimetric plate area; Based on the grid lines on the color chart, confirm the length of the branches and the growth stage of the citrus; Preprocess the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image; Construct a training set based on the corrected image, branch length, and growth stage information; A neural network model was built based on deep learning technology and trained using a training set until the network converged, resulting in an evaluation model for citrus shoot and leaf development. The evaluation model of citrus branch and leaf development is used to perform state recognition on the corrected image of the branch to be tested, obtain the current branch length, and determine whether the current leaf color is within a reasonable range.

2. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: Fix the colorimetric plate below the citrus branches, specifically: Use a fixed bracket with an adjustable buckle, clamp one end of the fixed bracket to the citrus tree trunk with the buckle, install a horizontally placed flat plate on the other end, and magnetically fix the colorimetric plate to the flat plate so that the colorimetric plate is directly below the citrus branch tip and remains level with the ground.

3. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: The colorimetric plate is a rectangular flat plate structure with its surface divided into two areas; one area is an evenly spaced grid line area, where the grid lines are squares with black lines, which are used to assist in confirming the length of branches; the other area is a standard color block area, where the color of the color block covers the color range of citrus leaves in different health conditions, which is used to perform color calibration on the collected images.

4. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: Determining the length of the shoots and the growth stage of the citrus fruits involves the following steps: The canny edge detection algorithm is used to identify the edges of the colorimetric plate grid lines and the outline edges of the citrus branches. The mutually perpendicular grid lines on the colorimetric plate are mapped to the parameter space through Hough transform to construct the grid line coordinate system; According to the position of the citrus branch in the grid coordinate system, the pixel length of the citrus branch in the image is calculated; Using the ratio between the side length of the grid lines on the color chart and the image pixels, the pixel length is converted into the actual physical length to obtain the branch length. The length of the branches and shoots was compared with the length range of the corresponding varieties in the pre-established citrus growth database to determine the growth stage of the citrus.

5. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: Preprocess the image information, specifically: Gaussian filtering is used to process the image to retain the main edge and structural information of the image; The image is processed by histogram equalization method to highlight the image details; The region of interest (ROI) in the image, including the citrus branches and the colorimetric plate, was determined and cropped, and the cropped image was scaled to an appropriate size using bilinear interpolation.

6. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: Color calibration based on the standard color blocks on the colorimetric chart includes the following steps: Segment the standard color block area of ​​the colorimetric plate from the preprocessed image, and extract the standard color block image using the image segmentation algorithm; Obtain the actual color value of each standard color block and compare it with the known standard color value, and calculate the color deviation using the color difference formula; The color value of each pixel in the image is corrected based on the color deviation, and the color is calculated and updated pixel by pixel to generate a corrected image.

7. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: The specific way to construct the training set is: Convert the corrected images into RGB format and assign a unique identifier to each corrected image using the image file name; Record the shoot length in numerical form, convert the growth stage information into category labels, and establish a correspondence with the corresponding image identifiers; Create an image folder to store the corrected images, and record the correspondence between image identifiers, branch length values, and growth stage labels in CSV format.

8. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: The neural network model adopts a multi-task parallel network architecture, including: feature extraction network, branch length prediction branch, and leaf color state classification branch; The feature extraction network uses the ResNet18 network to extract features from the input rectified image; The branch length prediction branch is used to extract spatial features related to branch length through the convolution layer and map the spatial features into predicted values ​​of branch length through the regression layer; The leaf color state classification branch is used to extract features related to the leaf color state through convolutional layers and pooling layers, and use the softmax activation function to output the probability of each leaf color state category.

9. The method for detecting citrus branch length and leaf color based on colorimetric plate correction according to claim 1, characterized in that: The specific steps of model training are: Set the learning rate, batch size, and number of training rounds, input the training set into the neural network model in batches, and obtain the predicted branch length value and leaf color state predicted probability distribution; Calculate the mean square error loss of branch length prediction and the cross entropy loss of leaf color state classification respectively, and sum them according to the corresponding weights to get the total loss; Calculate the gradient of the total loss with respect to the model parameters, and use the optimizer to update the model parameters according to the gradient; After completing all training rounds, the model parameters with the best performance were saved to obtain the evaluation model of citrus shoot and leaf development.

10. A citrus branch length and leaf color status detection system based on colorimetric plate correction, characterized in that: The method for detecting citrus branch length and leaf color status based on colorimetric plate correction according to any one of claims 1 to 9 comprises: The image acquisition module is used to fix the colorimetric plate below the citrus branches and use the mobile phone to synchronously capture image information including the complete branch area and the colorimetric plate area; The data acquisition module is used to confirm the length of branches and the growth stage of citrus fruits through the grid lines on the colorimetric plate; An image processing module is used to pre-process the image information and perform color calibration based on the standard color blocks on the colorimetric plate to generate a corrected image; The set construction module is used to construct the training set based on the corrected image, branch length and growth stage information; The model building and training module is used to build a neural network model using deep learning technology and train the model using a training set until the network converges to obtain an evaluation model for citrus shoot and leaf development; The real-time detection module is used to use the evaluation model of citrus branch and leaf development to perform state recognition on the corrected image of the branch to be tested, obtain the current branch length and determine whether the current leaf color is within a reasonable range.

Images