Image recognition optimization method based on deep learning algorithm

Abstract

The invention relates to the technical field of image recognition, and particularly discloses an image recognition optimization method based on a deep learning algorithm, and the method comprises the steps: 1, selecting a training data set for image recognition optimization, and carrying out the transformation of the training data, and enabling a model to be able to learn better robust features; 2, constructing a model of an image recognition task; according to the method, a deep learning model generally has relatively strong dependence on a large amount of annotated data by utilizing the matching effect of a plurality of steps, so that the problem of data dependence is solved, which means that the model can more effectively utilize limited data to learn and predict, the adaptability of the model in different data sets and scenes is improved, and the method is suitable for large-scale popularization and application. Moreover, the generalization ability is one of important indexes for measuring the performance of the deep learning model, and the improvement of the generalization ability of the model means that the model can better adapt to new data and scenes, thereby improving the accuracy and robustness of image recognition, and improving the application of the image recognition technology in wider scenes.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to an image recognition optimization method based on a deep learning algorithm. Background Art

[0002] Image recognition refers to the technology of using computers to process, analyze and understand images to identify targets and objects of various different patterns. It is a practical application of deep learning algorithms and an important field of artificial intelligence. Image recognition may be based on the main features of the image. Each image has its own characteristics, such as the letter A has a point, P has a circle, and the center of Y has an acute angle. In the process of image recognition, the perceptual mechanism will exclude the redundant input information and extract the key information. At the same time, there must be a mechanism in the brain responsible for integrating information, which can organize the information obtained in stages into a complete perceptual image.

[0003] Graphics recognition based on learning algorithms, especially image recognition technology based on deep learning, has made significant progress in recent years and has been widely used in many fields. Graphics recognition based on learning algorithms refers to the process of analyzing, understanding and identifying images using computer vision technology and machine learning algorithms. Its core is to enable computer systems to automatically extract features from images through learning algorithms, and classify and identify images based on these features. Deep learning algorithms, especially convolutional neural networks (CNNs), have made breakthrough progress in the field of image recognition and have become one of the current mainstream technologies.

[0004] The performance of image recognition algorithms depends largely on the quality of training data. If the training data is insufficient, biased or unbalanced, the recognition effect of the algorithm may be affected. At the same time, image recognition algorithms show high accuracy in specific fields and scenarios, but their generalization ability is still limited. When faced with some new and unseen images, the algorithm may make recognition errors, which limits the application of image recognition technology in a wider range of scenarios. To this end, an image recognition optimization method based on deep learning algorithm is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an image recognition optimization method based on a deep learning algorithm to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An image recognition optimization method based on a deep learning algorithm, comprising:

[0008] Step 1: Select a training data set optimized for image recognition and transform the training data so that the model can learn better robust features;

[0009] The image recognition training datasets include CI FAR-10 dataset, ImageNet dataset, MN IST dataset, VOC dataset and COCO dataset;

[0010] The transformation of the training data includes rotation, scaling, cropping and color transformation;

[0011] From the above, it can be seen that it is necessary to explain that the CI FAR-10 dataset is a commonly used color image dataset, which contains 60,000 32x32 color images, divided into 10 categories, each category has 6,000 images, of which 50,000 images are used for training and 10,000 images are used for testing. This dataset is very suitable for introductory exercises and algorithm verification for image classification;

[0012] The ImageNet dataset is a large visual database used for research and development of visual object recognition software. It contains millions of images organized into a hierarchy of categories, ranging from general categories (such as "animals") to very specific categories (such as "Dalmatian dogs"). The ImageNet dataset has an important position in the field of image recognition and computer vision, and is often used to train and evaluate deep learning models.

[0013] The MN I ST dataset is a handwritten digit dataset that contains a large number of handwritten digit images. Each image is a 28x28 grayscale image. This dataset is very suitable for training and testing digit recognition tasks.

[0014] The VOC dataset is an influential competition in the field of classification, detection, and segmentation of visual objects. Its dataset contains a variety of images and annotation information, which is very suitable for tasks such as image recognition, object detection, and image segmentation.

[0015] The COCO dataset is a large image dataset used for tasks such as object detection, segmentation, and image description. It contains rich images and detailed annotation information, and is very suitable for image recognition tasks in complex scenes.

[0016] By setting the above technical solution, in image recognition based on deep learning algorithm, a training data set optimized for image recognition is selected, and the training data is transformed so that the model can learn better robust features, which has the following benefits:

[0017] The optimized training dataset is usually of higher quality and representativeness, and can more comprehensively cover various scenarios and situations of image recognition, which enables the model to learn more accurate and rich features during the training process, thereby improving the accuracy of image recognition;

[0018] By selecting a training dataset that contains a variety of different scenes and situations, the model can learn more generalized features and thus better adapt to unknown or new image data;

[0019] The optimized training dataset is usually more compact and efficient, which can reduce unnecessary calculations and resource waste, help reduce model training time and cost, and improve overall development efficiency;

[0020] Transforming the training data has the following benefits:

[0021] By transforming the training data (such as rotation, cropping, color change, etc.), more equivalent data can be generated, thereby increasing the diversity of data. This diversity helps the model learn more robust features and improve its adaptability to new data.

[0022] Data transformation can increase the complexity of training data, making it more difficult for the model to overfit during training. Overfitting is a common problem in deep learning, that is, the model performs well on the training data but poorly on new data. Data transformation can effectively alleviate this problem.

[0023] The transformed training data can simulate more real-world scenarios and situations, making the model more robust when facing challenges such as noise, illumination changes, and occlusion. This robustness is of great significance for the practical application of image recognition systems.

[0024] In summary, combining the selection and optimization of the training data set with the transformation of the training data has the following benefits:

[0025] By optimizing the training data set and performing data transformation, the accuracy, generalization ability and robustness of the image recognition system can be significantly improved, so that it can show better performance in practical applications;

[0026] Optimizing training data sets and performing data transformation can reduce model training time and cost, while improving model stability and reliability, thereby reducing the overall cost of model development and maintenance.

[0027] Step 2: Build a model for the image recognition task and use the built model to automatically learn the feature representation of the image to capture the key information in the image;

[0028] The constructed models include convolutional neural network (CNN), recurrent neural network (RNN), ResNet, DenseNet and Inception;

[0029] The constructed model is also used to evaluate image recognition. The constructed model evaluates the performance of the image recognition algorithm through a validation set and a test set. The constructed model evaluation indicators include accuracy, recall and F1 score. For the hyperparameter adjustment of the constructed model, the optimal hyperparameter combination is found through random search or Bayesian optimization method.

[0030] As can be seen above, in image recognition tasks based on deep learning algorithms, building a model to automatically learn the feature representation of images and capture the key information in the image has the following benefits:

[0031] Traditional image recognition methods often rely on complex manual feature engineering, which requires experts to design feature extractors based on experience and knowledge. Deep learning models can automatically learn features from data, greatly reducing the need for manual feature engineering.

[0032] Through multi-layer convolution and pooling operations, deep learning models can efficiently capture key information in images, such as the shape, color, and texture of objects, thereby achieving accurate image recognition. Because deep learning models can learn robust feature representations from large amounts of data, they are highly resistant to noise and interference in images and can maintain high recognition accuracy in complex environments.

[0033] High accuracy: The deep learning model can capture more image details and key information by automatically learning the feature representation of the image, thereby improving the accuracy of recognition. Since the deep learning model can learn rich feature representations from a large amount of data, it has a strong generalization ability for new data and can maintain good recognition performance in unknown scenes.

[0034] The design process of deep learning models is relatively simple. You only need to determine the model's architecture and parameters, and then use a large amount of data for training. This greatly reduces the difficulty and complexity of model design. Deep learning models are highly scalable and optimizable, and performance can be further improved by adjusting the model's architecture, parameters, and training methods.

[0035] Step 3: Use the labeled data to supervise the construction model and optimize the network weights through the back propagation algorithm to minimize the loss function between the predicted results and the true labels;

[0036] The loss functions include cross entropy loss, mean square error loss, adversarial loss and center loss.

[0037] From the above, we can see that in image recognition based on deep learning algorithms, using labeled data to supervise the construction model and optimizing the network weights through the back propagation algorithm to minimize the loss function between the predicted results and the true labels has the following benefits:

[0038] Supervised learning uses pre-labeled datasets for training. These datasets contain images and their corresponding labels (such as object categories, bounding boxes, etc.); through training, the model can learn the mapping relationship between features in the image and labels, thereby improving accuracy and reliability on the test dataset;

[0039] The diversity and richness of the annotated dataset helps the model learn more image features and patterns; it enables the model to make accurate predictions and classifications when faced with unseen images, thus enhancing the model’s generalization ability.

[0040] The back-propagation algorithm calculates the gradient according to the loss function and propagates the error back to each layer to update the weights and biases. Through iterative optimization of multiple training cycles, the model can gradually reduce the prediction error and improve the recognition performance.

[0041] The back propagation algorithm is combined with optimization algorithms such as gradient descent to efficiently calculate the gradient of the loss function with respect to the network parameters; this enables the model to quickly adjust weights and biases during training, speeding up the training process and saving computing resources;

[0042] By minimizing the loss function, the performance of the model can be intuitively evaluated and used as a basis for model optimization and adjustment; the combination of supervised learning and back-propagation algorithm enables the model to quickly and accurately identify objects and scenes in images; at the same time, this deep learning-based image recognition technology not only improves recognition accuracy and efficiency, but also promotes the overall progress of artificial intelligence technology.

[0043] Step 4: Use the pre-trained model for transfer learning, use all the layers of the pre-trained model as the initial structure, and fine-tune the dataset for the specific task;

[0044] Step 5: Use the loss function to measure the performance of the model and design the loss function to guide model learning;

[0045] Step 6: Introduce the attention module in image recognition so that the constructed model can focus on important feature areas;

[0046] Step 7: Use visualization technology to display the key feature areas and decision-making basis in the image recognition process.

[0047] Wherein, the visualization techniques include feature map visualization and class activation mapping;

[0048] From the above, we can see that in image recognition based on deep learning algorithms, using visualization technology to display key feature areas and decision-making basis in the image recognition process has the following benefits:

[0049] Visualization technology can help researchers better understand the nature of deep features. For example, by visualizing feature vectors, the differences and similarities between different feature vectors can be found, thereby improving the accuracy of image classification.

[0050] Deep learning algorithms extract image features step by step through a multi-layer neural network structure to form a hierarchical feature representation. Visualization technology can display these hierarchical features, allowing researchers to have a deeper understanding of the image, which helps optimize the algorithm and improve recognition performance.

[0051] During the model training process, visualization technology can show the difference between the model's predicted results and the actual results, thereby guiding researchers to optimize and adjust the model; through the visualization analysis of different network structures, the advantages and disadvantages of different network structures in image classification can be explored, and a better network structure can be selected;

[0052] Deep learning models usually have a black box nature, and their decision-making process is difficult for humans to explain. Through visualization technology, the key feature areas and decision paths that the model focuses on when recognizing images can be displayed, thereby enhancing the interpretability of the model, helping researchers understand the decision-making process of the model, verify the rationality of the model, and discover potential errors and deviations.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention utilizes the cooperation between the above steps.

[0055] Solving data dependency issues has the following benefits:

[0056] Deep learning models usually have a strong dependence on a large amount of labeled data. Solving the data dependency problem means that the model can more effectively use limited data for learning and prediction, improving the adaptability of the model in different data sets and scenarios. Through technologies such as transfer learning, models pre-trained on large-scale data sets can be used to assist small sample learning, thereby reducing the cost of acquiring and labeling large-scale image data. After solving the data dependency problem, researchers can focus more on the innovation of model architecture and algorithms instead of spending a lot of time collecting and processing data.

[0057] Solving the problem of limited generalization has the following benefits:

[0058] Generalization ability is one of the important indicators to measure the performance of deep learning models. Improving the generalization ability of the model means that the model can better adapt to new data and scenarios, thereby improving the accuracy and robustness of image recognition. Models with strong generalization ability can make more reasonable predictions and judgments when facing unknown data, thereby enhancing the reliability of the model. Models with strong generalization ability can be applied to more diverse scenarios and tasks, such as autonomous driving, medical imaging diagnosis, etc., thereby promoting the application and development of deep learning technology in more fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the steps of the present invention. DETAILED DESCRIPTION

[0060] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] An image recognition optimization method based on a deep learning algorithm, comprising:

[0062] Step 1: Select a training data set optimized for image recognition and transform the training data so that the model can learn better robust features;

[0063] Step 2: Build a model for the image recognition task and use the built model to automatically learn the feature representation of the image to capture the key information in the image;

[0064] Step 3: Use the labeled data to supervise the construction model and optimize the network weights through the back propagation algorithm to minimize the loss function between the predicted results and the true labels;

[0065] Step 4: Use the pre-trained model for transfer learning, use all the layers of the pre-trained model as the initial structure, and fine-tune the dataset for the specific task;

[0066] Step 5: Use the loss function to measure the performance of the model and design the loss function to guide model learning;

[0067] Step 6: Introduce the attention module in image recognition so that the constructed model can focus on important feature areas;

[0068] Step 7: Use visualization technology to display the key feature areas and decision-making basis in the image recognition process.

[0069] The present invention utilizes the cooperation between the above steps.

[0070] Solving data dependency issues has the following benefits:

[0071] Deep learning models usually have a strong dependence on a large amount of labeled data. Solving the data dependency problem means that the model can more effectively use limited data for learning and prediction, improving the adaptability of the model in different data sets and scenarios. Through technologies such as transfer learning, models pre-trained on large-scale data sets can be used to assist small sample learning, thereby reducing the cost of acquiring and labeling large-scale image data. After solving the data dependency problem, researchers can focus more on the innovation of model architecture and algorithms instead of spending a lot of time collecting and processing data.

[0072] Solving the problem of limited generalization has the following benefits:

[0073] Generalization ability is one of the important indicators to measure the performance of deep learning models. Improving the generalization ability of the model means that the model can better adapt to new data and scenarios, thereby improving the accuracy and robustness of image recognition. Models with strong generalization ability can make more reasonable predictions and judgments when facing unknown data, thereby enhancing the reliability of the model. Models with strong generalization ability can be applied to more diverse scenarios and tasks, such as autonomous driving, medical imaging diagnosis, etc., thereby promoting the application and development of deep learning technology in more fields.

[0074] The image recognition training data sets include CI FAR-10 data set, ImageNet data set, MNIST data set, VOC data set and COCO data set;

[0075] The transformation of the training data includes rotation, scaling, cropping and color transformation;

[0076] As can be seen from the above, it should be noted that the CIFAR-10 dataset is a commonly used color image dataset, which contains 60,000 32x32 color images, divided into 10 categories, each category has 6,000 images, of which 50,000 images are used for training and 10,000 images are used for testing. This dataset is very suitable for introductory exercises and algorithm verification for image classification;

[0077] The ImageNet dataset is a large visual database used for research and development of visual object recognition software. It contains millions of images organized into a hierarchy of categories, ranging from general categories (such as "animals") to very specific categories (such as "Dalmatian dogs"). The ImageNet dataset has an important position in the field of image recognition and computer vision, and is often used to train and evaluate deep learning models.

[0078] The MNIST dataset is a handwritten digit dataset that contains a large number of handwritten digit images. Each image is a 28x28 grayscale image. This dataset is very suitable for training and testing digit recognition tasks.

[0079] The VOC dataset is an influential competition in the field of classification, detection, and segmentation of visual objects. Its dataset contains a variety of images and annotation information, which is very suitable for tasks such as image recognition, object detection, and image segmentation.

[0080] The COCO dataset is a large image dataset used for tasks such as object detection, segmentation, and image description. It contains rich images and detailed annotation information, and is very suitable for image recognition tasks in complex scenes.

[0081] By setting the above technical solution, in image recognition based on deep learning algorithm, a training data set optimized for image recognition is selected, and the training data is transformed so that the model can learn better robust features, which has the following benefits:

[0082] The optimized training dataset is usually of higher quality and representativeness, and can more comprehensively cover various scenarios and situations of image recognition, which enables the model to learn more accurate and rich features during the training process, thereby improving the accuracy of image recognition;

[0083] By selecting a training dataset that contains a variety of different scenes and situations, the model can learn more generalized features and thus better adapt to unknown or new image data;

[0084] The optimized training dataset is usually more compact and efficient, which can reduce unnecessary calculations and resource waste, help reduce model training time and cost, and improve overall development efficiency;

[0085] Transforming the training data has the following benefits:

[0086] By transforming the training data (such as rotation, cropping, color change, etc.), more equivalent data can be generated, thereby increasing the diversity of data. This diversity helps the model learn more robust features and improve its adaptability to new data.

[0087] Data transformation can increase the complexity of training data, making it more difficult for the model to overfit during training. Overfitting is a common problem in deep learning, that is, the model performs well on the training data but poorly on new data. Data transformation can effectively alleviate this problem.

[0088] The transformed training data can simulate more real-world scenarios and situations, making the model more robust when facing challenges such as noise, illumination changes, and occlusion. This robustness is of great significance for the practical application of image recognition systems.

[0089] In summary, combining the selection and optimization of the training data set with the transformation of the training data has the following benefits:

[0090] By optimizing the training data set and performing data transformation, the accuracy, generalization ability and robustness of the image recognition system can be significantly improved, so that it can show better performance in practical applications;

[0091] Optimizing training data sets and performing data transformation can reduce model training time and cost, while improving model stability and reliability, thereby reducing the overall cost of model development and maintenance.

[0092] The constructed models include convolutional neural network (CNN), recurrent neural network (RNN), ResNet, DenseNet and Inception;

[0093] The constructed model is also used to evaluate image recognition. The constructed model evaluates the performance of the image recognition algorithm through a validation set and a test set. The constructed model evaluation indicators include accuracy, recall and F1 score. For the hyperparameter adjustment of the constructed model, the optimal hyperparameter combination is found through random search or Bayesian optimization method.

[0094] As can be seen above, in image recognition tasks based on deep learning algorithms, building a model to automatically learn the feature representation of images and capture the key information in the image has the following benefits:

[0095] Traditional image recognition methods often rely on complex manual feature engineering, which requires experts to design feature extractors based on experience and knowledge. Deep learning models can automatically learn features from data, greatly reducing the need for manual feature engineering.

[0096] Through multi-layer convolution and pooling operations, deep learning models can efficiently capture key information in images, such as the shape, color, and texture of objects, thereby achieving accurate image recognition. Because deep learning models can learn robust feature representations from large amounts of data, they are highly resistant to noise and interference in images and can maintain high recognition accuracy in complex environments.

[0097] High accuracy: The deep learning model can capture more image details and key information by automatically learning the feature representation of the image, thereby improving the accuracy of recognition. Since the deep learning model can learn rich feature representations from a large amount of data, it has a strong generalization ability for new data and can maintain good recognition performance in unknown scenes.

[0098] The design process of deep learning models is relatively simple. You only need to determine the model's architecture and parameters, and then use a large amount of data for training. This greatly reduces the difficulty and complexity of model design. Deep learning models are highly scalable and optimizable, and performance can be further improved by adjusting the model's architecture, parameters, and training methods.

[0099] The loss functions include cross entropy loss, mean square error loss, adversarial loss and center loss.

[0100] From the above, we can see that in image recognition based on deep learning algorithms, using labeled data to supervise the construction model and optimizing the network weights through the back propagation algorithm to minimize the loss function between the predicted results and the true labels has the following benefits:

[0101] Supervised learning uses pre-labeled datasets for training. These datasets contain images and their corresponding labels (such as object categories, bounding boxes, etc.); through training, the model can learn the mapping relationship between features in the image and labels, thereby improving accuracy and reliability on the test dataset;

[0102] The diversity and richness of the annotated dataset helps the model learn more image features and patterns; it enables the model to make accurate predictions and classifications when faced with unseen images, thus enhancing the model’s generalization ability.

[0103] The back-propagation algorithm calculates the gradient according to the loss function and propagates the error back to each layer to update the weights and biases. Through iterative optimization of multiple training cycles, the model can gradually reduce the prediction error and improve the recognition performance.

[0104] The back propagation algorithm is combined with optimization algorithms such as gradient descent to efficiently calculate the gradient of the loss function with respect to the network parameters; this enables the model to quickly adjust weights and biases during training, speeding up the training process and saving computing resources;

[0105] By minimizing the loss function, the performance of the model can be intuitively evaluated and used as a basis for model optimization and adjustment; the combination of supervised learning and back-propagation algorithm enables the model to quickly and accurately identify objects and scenes in images; at the same time, this deep learning-based image recognition technology not only improves recognition accuracy and efficiency, but also promotes the overall progress of artificial intelligence technology.

[0106] Wherein, the visualization techniques include feature map visualization and class activation mapping;

[0107] From the above, we can see that in image recognition based on deep learning algorithms, using visualization technology to display key feature areas and decision-making basis in the image recognition process has the following benefits:

[0108] Visualization technology can help researchers better understand the nature of deep features. For example, by visualizing feature vectors, the differences and similarities between different feature vectors can be found, thereby improving the accuracy of image classification.

[0109] Deep learning algorithms extract image features step by step through a multi-layer neural network structure to form a hierarchical feature representation. Visualization technology can display these hierarchical features, allowing researchers to have a deeper understanding of the image, which helps optimize the algorithm and improve recognition performance.

[0110] During the model training process, visualization technology can show the difference between the model's predicted results and the actual results, thereby guiding researchers to optimize and adjust the model; through the visualization analysis of different network structures, the advantages and disadvantages of different network structures in image classification can be explored, and a better network structure can be selected;

[0111] Deep learning models usually have a black box nature, and their decision-making process is difficult for humans to explain. Through visualization technology, the key feature areas and decision paths that the model focuses on when recognizing images can be displayed, thereby enhancing the interpretability of the model, helping researchers understand the decision-making process of the model, verify the rationality of the model, and discover potential errors and deviations.

[0112] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition optimization method based on a deep learning algorithm, characterized in that: include: Step 1: Select a training data set optimized for image recognition and transform the training data so that the model can learn better robust features; Step 2: Build a model for the image recognition task and use the built model to automatically learn the feature representation of the image to capture the key information in the image; Step 3: Use the labeled data to supervise the construction model and optimize the network weights through the back propagation algorithm to minimize the loss function between the predicted results and the true labels; Step 4: Use the pre-trained model for transfer learning, use all the layers of the pre-trained model as the initial structure, and fine-tune the dataset for the specific task; Step 5: Use the loss function to measure the performance of the model and design the loss function to guide model learning; Step 6: Introduce the attention module in image recognition so that the constructed model can focus on important feature areas; Step 7: Use visualization technology to display the key feature areas and decision-making basis in the image recognition process.

2. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The image recognition training datasets include CIFAR-10 dataset, ImageNet dataset, MNIST dataset, VOC dataset and COCO dataset.

3. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The constructed models include convolutional neural networks (CNN), recurrent neural networks (RNN), ResNet, DenseNet and Inception.

4. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The transformation of the training data includes rotation, scaling, cropping and color transformation.

5. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The loss functions include cross entropy loss, mean square error loss, adversarial loss and center loss.

6. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The visualization techniques include feature map visualization and class activation mapping.

7. The image recognition optimization method based on deep learning algorithm according to claim 1, characterized in that: The constructed model is also used to evaluate image recognition.

8. The image recognition optimization method based on deep learning algorithm according to claim 7, characterized in that: The constructed model is evaluated by using a validation set and a test set to evaluate the performance of the image recognition algorithm.

9. The image recognition optimization method based on deep learning algorithm according to claim 8, characterized in that: The evaluation indicators of the constructed model include accuracy, recall and F1 score.

10. The image recognition optimization method based on deep learning algorithm according to claim 9, characterized in that: For the hyperparameter adjustment of the constructed model, the optimal hyperparameter combination is found through random search or Bayesian optimization method.

Images