Strawberry plant feature recognition method and system based on deep learning

Abstract

The invention discloses a strawberry plant feature recognition method and system based on deep learning, and the method comprises the following steps: collecting strawberry image data, marking the data, and dividing the data into a training set, a verification set and a test set; constructing a strawberry plant feature recognition model based on deep learning by adopting a YOLOv10 model, inputting image data, and performing model training and testing; performing model fine tuning by setting performance indexes, screening data, expanding a data set and improving the flooding ability of the model; hyper-parameters are adjusted to improve model performance, model pruning and model quantification are adopted to reduce the size of the model, and recognition accuracy is improved. Accuracy and efficiency of strawberry plant feature recognition are remarkably improved, and monitoring and management of a large-scale planting field are supported.

Description

Technical Field

[0001] The present invention relates to a method and system for identifying features of strawberry plants, and in particular to a method and system for identifying features of strawberry plants based on deep learning. Background Art

[0002] With the rapid development of intelligent agriculture, the strawberry planting industry is also actively seeking technological innovation to improve production efficiency. Traditional strawberry plant management mainly relies on manual observation and empirical judgment, which is not only labor-intensive but also easily affected by subjective factors, resulting in low management efficiency. The introduction of computer vision technology has improved these problems to a certain extent. However, most of the early image processing technologies relied on basic morphological operations and simple color filtering, with limited recognition accuracy and environmental adaptability, and it was difficult to handle complex backgrounds and changing lighting conditions.

[0003] In recent years, the development of deep learning technology, especially convolutional neural networks, has provided new solutions for the analysis of strawberry plant characteristics. Deep learning has demonstrated powerful performance in processing image recognition tasks, and can automatically extract complex image features, greatly improving the accuracy and robustness of recognition. At present, existing technologies have attempted to use deep learning in the agricultural field, but they are mostly limited to basic crop recognition and lack of detailed analysis of specific crops such as strawberries at different growth stages. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a high-accuracy and high-efficiency method and system for identifying strawberry plant features based on deep learning.

[0005] Technical solution: A strawberry plant feature recognition method based on deep learning described in the present invention comprises the following steps:

[0006] (1) Collect strawberry image data and mark the data into a training set, a validation set, and a test set;

[0007] (2) The YOLOv10 model was used to build a strawberry plant feature recognition model based on deep learning, and image data was input for model training and testing;

[0008] (3) Fine-tune the model by setting performance indicators, filtering data, expanding the data set, improving the generalization ability of the model, and adjusting the hyperparameters to improve model performance, and using model pruning and model quantization to reduce the size of the model and improve recognition accuracy.

[0009] The step (1) obtains characteristic image data of flowers, fruits and stems of strawberry plants from static pictures, video streams and cameras to achieve real-time monitoring, and uses Make Sense to annotate the data and divide them into a training set, a validation set and a test set.

[0010] In the step (2), during the model training process, the training data is enhanced by rotation, scaling, cropping, and color transformation.

[0011] The step (2) comprises the following steps:

[0012] (21) The YOLOv10 model was used to construct a strawberry plant feature recognition model based on deep learning;

[0013] (22) inputting the training set and the validation set into the strawberry plant feature recognition model to perform model training;

[0014] (23) Determine the model training status based on the loss function decline curve;

[0015] (24) The test set is input into the trained strawberry plant feature recognition model to obtain the detection results.

[0016] The step (3) comprises the following steps:

[0017] (31) Set accuracy, recall, and mAP performance indicators, filter data, remove noise, duplicate, and irrelevant data, and apply attention mechanisms to enhance feature extraction capabilities;

[0018] (32) Determine the learning rate, batch size, and optimizer parameters, and systematically search and adjust hyperparameters using grid search, random search, and Bayesian optimization methods;

[0019] (33) Perform weight pruning, structured pruning and importance pruning;

[0020] (34) Using dynamic quantization and static quantization, the weights are quantized from float32 to int8, and the activation values ​​are quantized during the inference process;

[0021] The step (3) also includes combining the model prediction results in the training phase to accurately identify the characteristics of the strawberry plants through voting, weighted averaging and stacking methods.

[0022] During the model training process, the training loss and verification indicators are monitored in real time and the training strategy is adjusted, and the fine-tuned model weights are saved regularly to prevent training interruptions.

[0023] The strawberry plant feature recognition system based on deep learning of the present invention comprises:

[0024] Data acquisition and preprocessing module: collect characteristic images of strawberry plants and remove noise, duplication and irrelevant data;

[0025] Strawberry plant feature recognition module: identifies the flowers, fruits, and stems of strawberry plants, conducts model training, and analyzes the test results;

[0026] Strawberry plant feature recognition model improvement module: fine-tuning and parameter optimization of the model, model pruning and quantization;

[0027] History Record Module: By tracing back historical identification results, users can conduct subsequent data analysis and management, providing a basis for a comprehensive assessment of growth status.

[0028] A computer device described in the present invention includes one or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the program is executed by the processor, the steps of the strawberry plant feature identification method based on deep learning are implemented.

[0029] The computer-readable storage medium described in the present invention stores a computer program thereon, and when the computer program is executed by a processor, the steps of the strawberry plant feature recognition method based on deep learning are implemented.

[0030] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: using the YOLOv10 deep learning model to improve recognition accuracy, it can accurately capture the subtle features of strawberry plants, and reduce the occurrence of false detection and missed detection; supporting batch image processing and real-time camera input, making real-time monitoring and management in large-scale strawberry plantations more efficient, saving a lot of time and labor costs; the design of the recognition history recording function is convenient for users to perform subsequent data analysis and management, and it is convenient to trace historical recognition results, providing a basis for a comprehensive evaluation of the growth status; it can not only show wide application potential in strawberry planting, but also can be expanded to other plant feature recognition scenarios, providing technical support for intelligent agricultural management; the friendly user interface and simple parameter setting in the system enable users to easily operate without complex technical background, reduce the technical threshold, and greatly facilitate the use of farmers and managers. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The figure is a diagram of the overall architecture of the method and system described in the present invention. DETAILED DESCRIPTION

[0032] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0033] A method for identifying strawberry plant features based on deep learning, characterized in that it comprises the following steps:

[0034] (1) Collect strawberry image data and mark the data into a training set, a validation set, and a test set;

[0035] (2) The YOLOv10 model was used to build a strawberry plant feature recognition model based on deep learning, and image data was input for model training and testing;

[0036] (3) Fine-tune the model by setting performance indicators, filtering data, expanding the data set, improving the generalization ability of the model, and adjusting the hyperparameters to improve model performance, and using model pruning and model quantization to reduce the size of the model and improve recognition accuracy.

[0037] Each step is described in detail below.

[0038] (1) Static and dynamic image data of strawberries captured by cameras and surveillance cameras. Each image is annotated using the Make Sense annotation tool. The dataset is divided into three parts: training set, validation set, and test set. The dataset contains three categories of strawberry targets: flowers, fruits, and stems.

[0039] (2) After data preparation is completed, the YOLOv10 model is used to build a strawberry plant feature recognition model based on deep learning, input image data, perform model training, and set the number of training rounds and batch size.

[0040]

[0041] The training data is enhanced by rotation, scaling, cropping, color conversion, etc. to simulate more scenarios.

[0042] The training set and the validation set are input into the strawberry plant feature recognition model to perform model training; the model training status is judged according to the loss function descent curve; the test set is input into the trained strawberry plant feature recognition model to obtain the detection results.

[0043] (3) Fine-tune the model by setting performance indicators, filtering data, expanding the data set, improving the generalization ability of the model, and adjusting the hyperparameters to improve model performance, and using model pruning and model quantization to reduce the size of the model and improve recognition accuracy.

[0044] Model fine-tuning:

[0045] Set performance indicators such as accuracy, recall, mAP, etc., remove noise, duplication, and irrelevant data; apply attention mechanism to enhance feature extraction capabilities; use learning rate scheduling, gradient clipping, early stopping strategy and other techniques for training; select a suitable initial learning rate, SGD as the optimizer, and mean square error as the loss function; use the weights of the pre-trained model as the initial point and load it into the model architecture.

[0046] During model training, monitor the training loss and validation indicators in real time and adjust the training strategy, save the fine-tuned model weights regularly to prevent training interruptions, and retain the best model.

[0047] Parameter optimization:

[0048] Determine the best learning rate, batch size, optimizer and other hyperparameters through experiments. Use grid search, random search or Bayesian optimization methods to systematically search and adjust hyperparameters, apply Dropout and L1 / L2 regularization methods to prevent overfitting, and use early stopping strategy to stop training when the performance of the validation set no longer improves.

[0049] Model pruning:

[0050] Use weight pruning to remove individual weights whose absolute values ​​are less than a certain threshold; use structured pruning to remove entire convolution kernels, neurons, or channels; use importance pruning to decide on pruning objects based on the importance scores of weights or activation values.

[0051] Implement pruning:

[0052] Use the PyTorch framework to implement pruning operations, such as:

[0053]

[0054] Model Quantization:

[0055] Using dynamic quantization and static quantization, weights are pre-quantized from float32 to int8 before inference, and activation values ​​are quantized during inference;

[0056] Quantization-aware training (QAT): simulates quantization operations during training, allowing the model to adapt to the accuracy loss caused by quantization.

[0057] Implementation Quantification:

[0058] Implementing quantization using the PyTorch framework

[0059]

[0060] Model Fusion:

[0061] Model integration and stacking are selected as the model fusion method. The prediction results of multiple models are weighted averaged or voted to improve the robustness and performance of the model. The outputs of multiple models are used as features to train a new model for final prediction.

[0062] Implementation of Fusion:

[0063]

[0064] After identifying the characteristics of the strawberry plant, the test results are marked on the picture to help users clearly understand the identification situation.

[0065] A strawberry plant feature recognition system based on deep learning, comprising:

[0066] Data acquisition and preprocessing module: collect characteristic images of strawberry plants and remove noise, duplication and irrelevant data;

[0067] Strawberry plant feature recognition module: identifies the flowers, fruits, and stems of strawberry plants, conducts model training, and analyzes the test results;

[0068] Strawberry plant feature recognition model improvement module: fine-tuning and parameter optimization of the model, model pruning and quantization;

[0069] History Record Module: By tracing back historical identification results, users can conduct subsequent data analysis and management, providing a basis for a comprehensive assessment of growth status.

[0070] During the recognition process, the system is used to load the model and enter the main interface. The user selects the input source and the software calls the YOLOv10 model for image processing, identifies strawberry features and calculates confidence. The recognition results are updated in real time on the interface, and the user can choose whether to save the results based on their needs. After completing the recognition task, the user can view historical records or perform new recognition tasks.

Claims

1. A strawberry plant feature recognition method based on deep learning, characterized in that: The following steps are involved: (1) Collect strawberry image data and mark the data into a training set, a validation set, and a test set; (2) The YOLOv10 model was used to build a strawberry plant feature recognition model based on deep learning, and image data was input for model training and testing; (3) Fine-tune the model by setting performance indicators, filtering data, expanding the data set, improving the generalization ability of the model, and performing model fine-tuning; Hyperparameters are adjusted to improve model performance, and model pruning and model quantization are used to reduce the size of the model and improve recognition accuracy.

2. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that, The step (1) obtains characteristic image data of flowers, fruits and stems of strawberry plants from static pictures, video streams and cameras to achieve real-time monitoring, and uses Make Sense to annotate the data and divide them into a training set, a validation set and a test set.

3. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that: In the step (2), during the model training process, the training data is enhanced by rotation, scaling, cropping, and color transformation.

4. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that, The step (2) comprises the following steps: (21) The YOLOv10 model was used to construct a strawberry plant feature recognition model based on deep learning; (22) inputting the training set and the validation set into the strawberry plant feature recognition model to perform model training; (23) Determine the model training status based on the loss function decline curve; (24) The test set is input into the trained strawberry plant feature recognition model to obtain the detection results.

5. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that: The step (3) comprises the following steps: (31) Set accuracy, recall, and mAP performance indicators, filter data, remove noise, duplicate, and irrelevant data, and apply attention mechanisms to enhance feature extraction capabilities; (32) Determine the learning rate, batch size, and optimizer parameters, and systematically search and adjust hyperparameters using grid search, random search, and Bayesian optimization methods; (33) Perform weight pruning, structured pruning and importance pruning; (34) Using dynamic quantization and static quantization, the weights are quantized from float32 to int8, and the activation values ​​are quantized during inference.

6. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that: The step (3) also includes combining the model prediction results in the training phase to accurately identify the characteristics of the strawberry plants through voting, weighted averaging and stacking methods.

7. A strawberry plant feature recognition method based on deep learning according to claim 1, characterized in that: During the model training process, the training loss and verification indicators are monitored in real time and the training strategy is adjusted, and the fine-tuned model weights are saved regularly to prevent training interruptions.

8. A strawberry plant feature recognition system based on deep learning, characterized in that: include: Data acquisition and preprocessing module: collect characteristic images of strawberry plants and remove noise, duplication and irrelevant data; Strawberry plant feature recognition module: identifies the flowers, fruits, and stems of strawberry plants, conducts model training, and analyzes the test results; Strawberry plant feature recognition model improvement module: fine-tuning and parameter optimization of the model, model pruning and quantization; History Record Module: By tracing back historical identification results, users can conduct subsequent data analysis and management, providing a basis for a comprehensive assessment of growth status.

9. A computer device, characterized in that: The method comprises one or more processors, a memory and one or more programs, wherein the one or more programs are stored in the memory and are configured to be executed by the one or more processors, and when the programs are executed by the processors, the steps of a strawberry plant feature recognition method based on deep learning are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a strawberry plant feature recognition method based on deep learning as described in any one of claims 1 to 7 are implemented.

Images