Intelligent measurement method for stem diameter of rice

Abstract

The application discloses a kind of rice stem diameter intelligent measurement methods.The method includes the following steps: setting reference, using shooting tool to collect the rice stem cross-sectional image data with reference, and the rice stem cross-sectional image data with reference is preprocessed, obtains image dataset;YOLOv8s model and Faster R-CNN model are used to build rice stem diameter intelligent measurement model;Image dataset is used to train, verify and test rice stem diameter intelligent measurement model, and the measurement result of rice stem diameter is obtained using the rice stem diameter intelligent measurement model after testing and the image dataset to be measured.The present application can improve the unreliability, generalization performance and precision of the measurement model, thereby improving the accuracy of the rice stem diameter intelligent measurement result, and reducing the difficulty and cost of obtaining the rice stem cross-sectional image to be measured, with stronger practical operability.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent measurement of rice stem diameter, and particularly relates to an intelligent measurement method of rice stem diameter. BACKGROUND

[0002] As a channel for supporting rice plants and transporting nutrients, the thickness of rice stem has an important influence on yield. Thick and strong stems can resist wind and disease invasion and reduce lodging. At the same time, good stem structure is also conducive to rapid nutrient transport and distribution, laying the foundation for high yield. The automatic measurement method of rice stem cross-sectional diameter can efficiently obtain a large amount of rice stem cross-sectional diameter data, helping breeders quickly select rice varieties with moderate diameter, which is conducive to lodging resistance and yield improvement. Compared with the traditional manual measurement method, the automatic measurement method of rice stem cross section can reduce human error and improve data accuracy, providing more reliable data support for breeding research.

[0003] The existing method introduces neural network research into rice stem recognition, making the automatic measurement of rice stem cross-sectional diameter more convenient. The existing method one uses a deep learning model to segment the rice cross-sectional CT image and calculate the corresponding microstructure parameters, and then completes the automatic measurement of the rice stem cross-sectional diameter. The existing method two constructs a rice stem slice image segmentation model based on an improved Mask R-CNN network for measuring and counting the number of vascular bundles and area in the rice stem microscopic slice image.

[0004] In summary, the current rice stem cross-sectional diameter measurement method mainly obtains images by taking rice stem slices as objects, and then uses a neural network model for prediction. The main defects are:

[0005] (1) The neural network prediction model used has unreliability, limited generalization performance and precision, which leads to inaccurate measurement results;

[0006] (2) The rice stem needs to be manually sliced and taken back to the indoor to take the stem cross-sectional image by CT or microscope, so the cost of stem cross-sectional CT image or microscopic image acquisition is high, and it is not convenient for users to use. SUMMARY

[0007] In view of the above deficiencies in the prior art, the present application provides an intelligent measurement method of rice stem diameter.

[0008] In order to achieve the above-mentioned application purposes, the technical scheme adopted by the present application is:

[0009] An intelligent measurement method of rice stem diameter, comprising the following steps:

[0010] S1, set a reference, use a shooting tool to collect rice stem cross-section image data with a reference, and pretreat the rice stem cross-section image data with the reference to obtain an image data set;

[0011] S2, construct a rice stem diameter intelligent measurement model based on a YOLOv8s model and a Faster R-CNN model;

[0012] S3, train, verify and test the rice stem diameter intelligent measurement model in step S2 using the image data set in step S1, collect an image data set to be measured, and use the tested rice stem diameter intelligent measurement model and the image data set to be measured to obtain a measurement result of the rice stem diameter.

[0013] Further, in step S1, the shooting tool is used to collect rice stem cross-section image data with a reference, and the specific process is as follows: cut off the mature rice panicle, move the cut-off rice panicle away from the rice stubble position, keep the rice stubble height at 10 cm, set the reference beside the rice stubble, and then use the shooting tool to collect the rice stem cross-section image at a position 15-30 cm above the rice stem cross-section perpendicular to the rice stem cross-section to obtain the rice stem cross-section image data with the reference.

[0014] Further, in step S1, the rice stem cross-section image data with the reference is pretreated to obtain an image data set, including the following steps:

[0015] A1, filter the rice stem cross-section image data with the reference using a median filtering method to obtain filtered rice stem cross-section image data with the reference;

[0016] A2, label the reference and the stem cross-section in the filtered rice stem cross-section image data with the reference in step A1 using a data labeling method to obtain rice stem cross-section image data containing label and position information;

[0017] A3, augment the rice stem cross-section image data containing label and position information in step A2 using a data augmentation method to obtain an image data set.

[0018] Further, step A2 includes the following steps:

[0019] A21, based on the filtered rice stem cross-section image data with the reference in step A1, draw a rectangular frame on the rice stem cross-section and the reference in the image respectively to obtain rice stem cross-section image data with a predicted bounding box;

[0020] A22. Based on the rice stalk cross-sectional image data with predicted bounding boxes from step A21, store the labels and coordinates of the predicted bounding boxes into a file with a set format to obtain rice stalk cross-sectional image data containing labels and location information.

[0021] Further, in step S2, the intelligent measurement model for rice stem diameter includes an input layer, a YOLOv8s model, a first non-maximum suppression layer, a Faster R-CNN model, a second non-maximum suppression layer, a weighted adjustment layer, and an output layer. The input layer is connected to both the input of the YOLOv8s model and the Faster R-CNN model. The outputs of the YOLOv8s model and the Faster R-CNN model are respectively connected to the inputs of the first and second non-maximum suppression layers. The outputs of both the first and second non-maximum suppression layers are connected to the input of the weighted adjustment layer, and the output of the weighted adjustment layer is connected to the output layer. The output layer is used to obtain the intelligent measurement results of rice stem diameter.

[0022] Furthermore, the YOLOv8s model is improved by adding an SENet module with an attention mechanism after the last convolutional layer of the YOLOv8s model. The SENet module is used to extract key information from the cross-sectional image data of rice stalks.

[0023] Furthermore, the Faster R-CNN model is improved by removing the commonly stacked convolutional layers and setting the ResNet50 model as the backbone network for extracting basic features from rice stalk cross-section images. A feature pyramid structure is then set after the ResNet50 model. The ResNet50 model is used to extract features from the image dataset to generate feature maps of rice stalk cross-sections. The feature pyramid structure is used to fuse low-resolution features with high-resolution features based on the feature maps of rice stalk cross-sections through top-down paths and lateral connections, generating multi-scale feature maps of rice stalk cross-sections.

[0024] Furthermore, the weighted adjustment layer sets the initial weight ratio of the output of the first nonmaximum suppression layer to the output of the second nonmaximum suppression layer to 1:3, and uses grid search to iterate the initial weight ratio.

[0025] Furthermore, the data processing procedure for obtaining the intelligent measurement results of rice stem diameter in the output layer is represented as follows:

[0026] ;

[0027] in: The actual value of the diameter of the rice stalk. Pixel diameter for reference, Max value symbol, Long side of the predicted bounding box of the rice stem cross section, Short side of the predicted bounding box of the rice stem cross section.

[0028] Further, in step S3, the rice stem diameter intelligent measurement model in step S2 is trained, verified and tested by using the image data set in step S1, and the specific process is as follows: the image data set in step S1 is randomly divided into a training set, a verification set and a test set according to a ratio of 0.6:0.2:0.2, and the rice stem diameter intelligent measurement model in step S2 is trained, verified and tested by using the training set, the verification set and the test set, respectively.

[0029] The beneficial effects of the present application are:

[0030] (1) The present application can improve the unreliability, generalization performance and precision of the measurement model by constructing a rice stem diameter intelligent measurement model based on a YOLOv8s model and a Faster R-CNN model, thereby improving the accuracy of the rice stem diameter intelligent measurement result.

[0031] (2) The present application can reduce the difficulty and cost of obtaining the rice stem cross section image to be measured by setting a reference object, using a shooting tool to collect the rice stem cross section image data with the reference object, and preprocessing the rice stem cross section image data with the reference object to obtain an image data set, which has stronger practical operability. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 It is a flowchart of a rice stem diameter intelligent measurement method;

[0033] Figure 2 It is a process diagram of collecting the rice stem cross section image with the reference object in the present application;

[0034] Figure 3 It is a structure diagram of the rice stem diameter intelligent measurement model in the present application;

[0035] Figure 4 It is a structure diagram of the improved YOLOv8s model in the present application;

[0036] Figure 5 It is a structure diagram of the SENet module;

[0037] Figure 6 It is a structure diagram of the improved Faster R-CNN model in the present application. DETAILED DESCRIPTION

[0038] The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, all the inventions utilizing the concept of the present application are within the scope of the present application as long as various changes are obvious within the spirit and scope of the present application defined and determined by the appended claims.

[0039] As shown in Figure 1 A method for intelligently measuring the diameter of rice stems, comprising steps S1-S3, specifically as follows:

[0040] S1, setting a reference object, using a shooting tool to collect rice stem cross-sectional image data with the reference object, and preprocessing the rice stem cross-sectional image data with the reference object to obtain an image data set.

[0041] In an optional embodiment of the present application, a circular iron block with a diameter of 3 cm is selected as the reference object.

[0042] As shown in Figure 2 The present application uses a shooting tool to collect rice stem cross-sectional image data with the reference object, and the specific process is as follows: cutting mature rice panicles, moving the cut rice panicles away from the rice stubble position, leaving the rice stubble height at 10 cm, and setting the reference object beside the rice stubble, then using a shooting tool to collect rice stem cross-sectional images at a position 15-30 cm above the rice stem cross section perpendicular to the rice stem cross section in the rice field, to obtain rice stem cross-sectional image data with the reference object. The shooting tool in the present application is a smart phone.

[0043] The present application preprocesses the rice stem cross-sectional image data with the reference object to obtain an image data set, including the following steps:

[0044] A1, using a median filtering method to filter the rice stem cross-sectional image data with the reference object to obtain filtered rice stem cross-sectional image data with the reference object.

[0045] Specifically, the collected rice stem cross-sectional image data with the reference object in the present application contains different light, different complex background, and different weather conditions, and the image data is affected by different degrees of noise, therefore, the present application uses a median filtering method to filter the rice stem cross-sectional image data with the reference object to obtain filtered rice stem cross-sectional image data with the reference object.

[0046] A2, using a data labeling method to label the reference object and the stem cross section in the filtered rice stem cross-sectional image data with the reference object in step A1 to obtain rice stem cross-sectional image data containing label and position information.

[0047] Step A2 comprises the following steps:

[0048] A21, based on the rice stem cross-section image data with reference objects filtered in step A1, draw a rectangular frame on the rice stem cross-section and the reference objects in the image respectively, and obtain rice stem cross-section image data with a predicted bounding box.

[0049] A22, based on the rice stem cross-section image data with a predicted bounding box in step A21, store the label and coordinates of the predicted bounding box into a file in a specified format, to obtain rice stem cross-section image data containing label and position information.

[0050] A3, using data augmentation method to augment the rice stem cross-section image data containing label and position information in step A2, to obtain an image data set.

[0051] S2, based on the YOLOv8s model and the Faster R-CNN model, construct a rice stem diameter intelligent measurement model.

[0052] In an optional embodiment of the present application, the rice stem diameter intelligent measurement model comprises an input layer, a YOLOv8s model, a first non-maximum suppression layer, a Faster R-CNN model, a second non-maximum suppression layer, a weighted adjustment layer and an output layer.

[0053] As shown in Figure 3 , the input layer is connected to the input end of the YOLOv8s model and the input end of the Faster R-CNN model at the same time, the output end of the YOLOv8s model and the output end of the Faster R-CNN model are connected to the input end of the first non-maximum suppression layer and the input end of the second non-maximum suppression layer respectively, the output end of the first non-maximum suppression layer and the output end of the second non-maximum suppression layer are both connected to the input end of the weighted adjustment layer, and the output end of the weighted adjustment layer is connected to the output layer. The output layer is used to obtain the intelligent measurement result of the rice stem diameter.

[0054] Specifically, since the rice stem cross-section image is taken under complex conditions in the field, it will cause the pixels of the rice stem cross-section and some objects in the image background to be very similar, which will affect the accuracy of the rice stem diameter intelligent measurement model. Therefore, the present application improves the rice stem diameter intelligent measurement model to enhance the accuracy of the rice stem diameter intelligent measurement model.

[0055] The present application improves the YOLOv8s model, specifically: adding a SENet module introducing attention mechanism after the last convolution layer of the YOLOv8s model, the SENet module is used to extract the key information of the rice stem cross-section image data.

[0056] As Figure 4 shown, the present application provides an improved YOLOv8s model structure diagram, Conv is a convolutional layer, which is a standard convolutional operation module, used to process the input rice stem cross-section feature map through convolutional operation to extract the local features in the rice stem cross-section image; C2f is CSP Bottleneck, which is an improved bottleneck module composed of multiple DarkNet Bottleneck, used to extract and fuse the rice stem cross-section features; SPPF is SpatialPyramid Pooling-Fast, used for spatial pyramid pooling, which extracts rice stem cross-section features of different scales through multiple max pooling operations; Concat is a feature connection module, used to connect tensors along the specified dimension; Upsample is an up-sampling module, used to enlarge the low-resolution rice stem cross-section feature map to the highest resolution; Detect is a detection head, used to predict the rice stem cross-section bounding box and class probability; Conv2d is a two-dimensional convolution, used to capture the local features of the rice stem cross-section; BN is BatchNorm batch normalization, used to speed up the training process of the rice stem cross-section detection model; SiLU is Sigmoid Linear Unit activation function, used to handle complex nonlinear relationships and enhance the expression ability of the model; BottleNeck is a bottleneck layer, which is a residual structure used to enhance network depth and reduce parameter quantity; add is a tensor addition, used to add two tensors; MAXPool is max pooling, used to reduce the spatial size of the rice stem cross-section feature map; Split is a feature layer, used to split the input tensor into multiple tensors along the specified dimension; BboxLoss is the bounding box loss, used to calculate the difference between the predicted bounding box and the true bounding box; ClsLoss is the classification loss, used to calculate the difference between the predicted class and the true class.

[0057] As Figure 5 shown, the present application provides a structure diagram of SENet module, H is the height of the rice stem cross-section feature map, C is the number of channels of the rice stem cross-section feature map, and W is the width of the rice stem cross-section feature map; F tr is the process of performing convolution operation on the input feature map to obtain the intermediate feature map U; F sq is the process of compressing the intermediate feature map U into a 1x1xC vector through global average pooling; 1x1xC is the dimension of the rice stem cross-section feature map, 1x1 is the size of the rice stem cross-section feature map, and C is the number of channels of the rice stem cross-section feature map; F ex is the process of nonlinear transformation of the output of F sq by two fully connected layers, learning the dependency between channels, and outputting channel weights;scale To F ex The output channel weight is multiplied with the intermediate feature map U channel by channel, thereby adjusting the importance of each channel, and finally obtaining the enhanced rice stem cross-section feature map.

[0058] The application improves the Faster R-CNN model, specifically: removing the ordinary stacked convolutional layer in the Faster R-CNN model, setting the ResNet50 model as the basic feature extraction backbone network of the rice stem cross-section image, and setting the feature pyramid structure after the ResNet50 model; the ResNet50 model is used to extract the features of the image data set to generate the feature map of the rice stem cross-section; the feature pyramid structure is used to fuse the low resolution features and the high resolution together through the top-down path and the horizontal connection according to the feature map of the rice stem cross-section, to generate the multi-scale feature map of the rice stem cross-section.

[0059] As Figure 6As shown, the present application provides an improved Faster R-CNN model structure diagram, 0.5x is the scaling ratio of the model, used to adjust the size of the feature map; C1 is the output of the first layer of the feature extraction network, C2 is the output of the second layer of the feature extraction network; C3 is the output of the third layer of the feature extraction network, C4 is the output of the fourth layer of the feature extraction network; C5 is the output of the fifth layer of the feature extraction network; FPN is a feature pyramid network, used to process rice stem cross-section feature maps of different scales, and enhance the detection ability of the model for multi-scale rice stem cross-sections; 1x1 is a 1x1 convolutional layer, used to adjust the number of channels of the feature map; 3x3 is a 3x3 convolutional layer, used to extract more rice stem cross-section feature information; M2 is the second layer output of the feature pyramid network after 1x1 convolution; M3 is the third layer output of the feature pyramid network after 1x1 convolution; M4 is the fourth layer output of the feature pyramid network after 1x1 convolution; M5 is the fifth layer output of the feature pyramid network after 1x1 convolution; P2 is the second layer output of the feature pyramid network after 3x3 convolution, P3 is the third layer output of the feature pyramid network after 3x3 convolution, P4 is the fourth layer output of the feature pyramid network after 3x3 convolution, P5 is the fifth layer output of the feature pyramid network after 3x3 convolution, and P6 is the sixth layer output of the feature pyramid network after 3x3 convolution; RPN is a region proposal network, used to generate candidate regions; reg is a regression, used to correct the position of the candidate region, cls is a classification, used to determine the class of the candidate region, 512x512 proposals are candidate regions of 512x512 size, 256x256 proposals are candidate regions of 256x256 size, 128x128 proposals are candidate regions of 128x128 size, 64x64 proposals are candidate regions of 64x64 size, and 32x32 proposals are candidate regions of 32x32 size; 2xup is an up-sampling operation, used to adjust the size of the rice stem cross-section feature map; 1x1conv is a 1x1 convolutional layer, used to adjust the number of channels of the rice stem cross-section feature map; ROI is a region of interest, representing the region of interest extracted from the image; ROI Pooling is a region of interest pooling, used to convert ROIs of different sizes into fixed-size rice stem cross-section feature maps; FC is a fully connected layer, used to further process the rice stem cross-section feature map; Softmax Classify is a Softmax classification, used to determine the class of the candidate region; and BBox Regression is a bounding box regression, used to correct the position of the candidate region.

[0060] Specifically, because there are objects similar to the pixels and image background of the rice stem cross section in the present application, the YOLOv8s model and the Faster R-CNN model will generate many overlapping prediction bounding boxes with different confidence levels. The present application sets a first non-maximum suppression layer to suppress the non-maximum prediction bounding boxes in the YOLOv8s model, and sets a second non-maximum suppression layer to suppress the non-maximum prediction bounding boxes in the Faster R-CNN model, so as to reduce the number of prediction bounding boxes and retain the best prediction box.

[0061] The weighting adjustment layer sets the initial weight ratio of the output end of the first non-maximum suppression layer and the output end of the second non-maximum suppression layer to 1:3, and iterates the initial weight ratio using grid search.

[0062] The output layer obtains the data processing process of the intelligent measurement result of the rice stem diameter, which is represented as:

[0063] ;

[0064] Wherein: is the actual value of the measured diameter of the rice stem, is the pixel diameter of the reference object, is the maximum value symbol, is the long side of the rice stem cross section prediction bounding box, is the short side of the rice stem cross section prediction bounding box.

[0065] The present application calculates the pixel diameter of the reference object, which is represented as:

[0066] ;

[0067] Wherein: is the pixel diameter of the reference object, is the long side of the reference object prediction bounding box, is the short side of the reference object prediction bounding box.

[0068] S3, using the image data set in step S1 to train, verify and test the intelligent measurement model of the rice stem diameter in step S2, collecting the image data set to be measured, and using the tested intelligent measurement model of the rice stem diameter and the image data set to be measured to obtain the measurement result of the rice stem diameter.

[0069] In an optional embodiment of the present application, the present application trains, verifies and tests the rice stem diameter intelligent measurement model in step S2 using the image data set in step S1, and the specific process is as follows: the image data set in step S1 is randomly divided into a training set, a verification set and a test set according to a ratio of 0.6:0.2:0.2, and the rice stem diameter intelligent measurement model in step S2 is trained, verified and tested using the training set, the verification set and the test set respectively.

[0070] The present application simulates the rice stem diameter intelligent measurement model in the present method, and the simulation results are shown in Table 1:

[0071]

[0072] As can be seen from Table 1, the recall rate and F1 value of the rice stem diameter intelligent measurement model provided by the present application are higher than those of the YOLOv8s model and the Faster R-CNN model, which indicates that the rice stem diameter intelligent measurement model provided by the present application can more accurately detect the cross section of the rice stem, thereby providing a solid prerequisite for the intelligent measurement of the diameter of the rice stem.

[0073] Those skilled in the art will appreciate that the embodiments described herein are intended to facilitate the understanding of the principles of the present application and should be understood as not limiting the scope of protection of the present application to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations according to the technical inspirations disclosed in the present application without departing from the essence of the present application, and these modifications and combinations are still within the scope of protection of the present application.

Claims

1. A method for intelligent measurement of rice stalk diameter, characterized in that, Includes the following steps: S1. Set up a reference object, use a camera to collect cross-sectional image data of rice stalks with the reference object, and preprocess the cross-sectional image data of rice stalks with the reference object to obtain an image dataset; S2. Construct an intelligent measurement model for rice stem diameter based on the YOLOv8s model and the Faster R-CNN model; The intelligent rice stem diameter measurement model comprises an input layer, a YOLOv8s model, a first non-maximum suppression layer, a Faster R-CNN model, a second non-maximum suppression layer, a weighted adjustment layer, and an output layer. The input layer is connected to both the inputs of the YOLOv8s model and the Faster R-CNN model. The outputs of the YOLOv8s model and the Faster R-CNN model are connected to the inputs of the first and second non-maximum suppression layers, respectively. The outputs of both the first and second non-maximum suppression layers are connected to the input of the weighted adjustment layer, and the output of the weighted adjustment layer is connected to the output layer. The output layer is used to acquire the intelligent measurement results of the rice stem diameter. The YOLOv8s model is improved by adding an SENet module with an attention mechanism after the last convolutional layer of the YOLOv8s model. The SENet module is used to extract key information from the cross-sectional image data of rice stalks. The Faster R-CNN model is improved by removing the stacked convolutional layers and setting the ResNet50 model as the backbone network for feature extraction of rice stalk cross-section images. A feature pyramid structure is then placed after the ResNet50 model. The ResNet50 model is used to extract features from the image dataset to generate feature maps of rice stalk cross-sections. The feature pyramid structure is used to fuse low-resolution features with high-resolution features based on the feature maps of rice stalk cross-sections through top-down paths and lateral connections, generating multi-scale feature maps of rice stalk cross-sections. The data processing procedure for obtaining the intelligent measurement results of rice stem diameter from the output layer is represented as follows: ； in: The actual value of the diameter of the rice stalk. The pixel diameter of the reference object. To determine the sign of the maximum value, Predict the long side of the bounding box for the cross-section of rice stalks. Predict the short side of the bounding box for the cross-section of rice stalks; S3. Use the image dataset from step S1 to train, validate, and test the intelligent measurement model for rice stem diameter in step S2. Collect the image dataset to be measured, and use the tested intelligent measurement model for rice stem diameter and the image dataset to be measured to obtain the measurement results of rice stem diameter.

2. The intelligent measurement method for rice stem diameter according to claim 1, characterized in that, In step S1, a camera is used to collect cross-sectional image data of rice stalks with a reference object. The specific process is as follows: the mature rice ears are cut off, and the cut rice ears are moved away from the stubble position, leaving a stubble height of 10 cm. The reference object is set next to the stubble, and then the camera is used to collect cross-sectional images of rice stalks at a position 15-30 cm directly above the cross-section of the rice stalk in the paddy field to obtain cross-sectional image data of rice stalks with a reference object.

3. The intelligent measurement method for rice stalk diameter according to claim 1, characterized in that, In step S1, the cross-sectional image data of rice stalks with reference objects is preprocessed to obtain an image dataset, including the following steps: A1. Use median filtering to filter the cross-sectional image data of rice stalks with reference objects to obtain the filtered cross-sectional image data of rice stalks with reference objects. A2. Use the data annotation method to annotate the reference objects and stem cross-sections in the filtered rice stem cross-section image data with reference objects in step A1, and obtain rice stem cross-section image data containing labels and location information; A3. Use data augmentation to augment the cross-sectional image data of rice stalks containing labels and location information in step A2 to obtain an image dataset.

4. The intelligent measurement method for rice stalk diameter according to claim 3, characterized in that, Step A2 includes the following steps: A21. Based on the rice stalk cross-section image data with reference objects after filtering in step A1, draw rectangular boxes for the rice stalk cross-section and the reference objects in the image respectively to obtain rice stalk cross-section image data with predicted bounding boxes. A22. Based on the rice stalk cross-sectional image data with predicted bounding boxes from step A21, store the labels and coordinates of the predicted bounding boxes into a file with a set format to obtain rice stalk cross-sectional image data containing labels and location information.

5. The intelligent measurement method for rice stem diameter according to claim 1, characterized in that, The weighted adjustment layer sets the initial weight ratio of the output of the first nonmaximum suppression layer to the output of the second nonmaximum suppression layer to 1:3, and uses grid search to iterate the initial weight ratio.

6. The intelligent measurement method for rice stalk diameter according to claim 1, characterized in that, In step S3, the intelligent measurement model for rice stem diameter in step S2 is trained, validated, and tested using the image dataset from step S1. Specifically, the image dataset from step S1 is randomly divided into a training set, a validation set, and a test set in a ratio of 0.6:0.2:0.

2. The intelligent measurement model for rice stem diameter in step S2 is then trained, validated, and tested using the training set, validation set, and test set, respectively.

Images