Article automatic classification method and system based on artificial intelligence

Abstract

The invention provides an automatic article classification method and system based on artificial intelligence, and the method comprises the steps: collecting various types of image information of articles, carrying out the feature extraction of the various types of image information, obtaining main features, building an article information library based on the main features, obtaining the information features of the articles in various forms in advance, and carrying out the classification of the articles. A rich and comprehensive article information basis is provided for realizing automatic article classification, an RGB grouping attention mechanism is utilized to perform feature extraction and processing on main features in the article information base to obtain an effective feature image, an accurate and effective feature image is provided for training and establishment of a classification model, and based on the effective feature image, artificial intelligence is combined to obtain an accurate and effective classification model. And training the initial classification model to obtain the target classification model, inputting the collected image of the to-be-classified article into the target classification model, and determining the category of the to-be-classified article according to the output result, so that the classified article is quickly and accurately identified, the time and energy of manual classification are greatly reduced, and the working efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of item classification, and particularly to an automatic item classification method and system based on artificial intelligence. Background Art

[0002] In the retail industry, automatic item classification technology uses technologies such as image recognition and sensors to automatically identify the appearance and label information of goods and accurately classify them onto the corresponding shelves. This automated classification method not only improves classification efficiency, reduces human errors, but also helps retailers monitor the sales situation and inventory levels of goods in real time, ensuring that there is always an adequate supply of goods on the shelves. In addition, by analyzing consumers' purchase history and preferences, the automated technology can achieve personalized product recommendations, increasing sales and customer satisfaction.

[0003] During the automatic item classification process, due to the angle of the item, item extrusion, etc., the item cannot be well recognized and classified, and the accuracy of image recognition technology is insufficient. These situations result in the inability to achieve precise recognition of the item, affecting subsequent operations on the item. Summary of the Invention

[0004] The present invention provides an automatic item classification method and system based on artificial intelligence to solve the problems raised in the background art.

[0005] An automatic item classification method based on artificial intelligence includes:

[0006] S1: Collect various types of image information of items, extract main features from the various types of image information, and establish an item information library based on the main features;

[0007] S2: Use the RGB grouped attention mechanism to extract and process the main features in the item information library to obtain an effective feature image;

[0008] S3: Based on the effective feature image, combined with artificial intelligence, train an initial classification model to obtain a target classification model;

[0009] S4: Input the captured image of the item to be classified into the target classification model, and determine the category of the item to be classified according to the output result.

[0010] Preferably, in S1, collecting various types of image information of items includes:

[0011] Collect various types of image information of items from historical image collections;

[0012] Use a classification decision tree to classify the various types of image information, and obtain the image information corresponding to each item according to the classification result.

[0013] Preferably, in S1, the main features are extracted from the various types of image information, and an item information database is established based on the main features, including:

[0014] Preprocess the image information corresponding to each item to obtain standard image information;

[0015] Extract the main features from the standard image information;

[0016] An item information database is established based on the correspondence between the item category of each item and the main features.

[0017] Preferably, it further includes: updating the item information database in real time, specifically:

[0018] Obtain the captured images of historical items to be classified in the most recent time period and their corresponding classification result information;

[0019] Match the captured images of the historical items to be classified and their corresponding classification result information with the item information database. If a corresponding item is matched, update the information of the corresponding item based on the captured images of the historical items to be classified and their corresponding classification result information. If no corresponding item is matched, establish new item information in the item information database based on the captured images of the historical items to be classified and their corresponding classification result information.

[0020] Preferably, in S2, the RGB group attention mechanism is used to extract and process the main features in the item information database to obtain an effective feature image, including:

[0021] Use the RGB group attention mechanism to extract global and local features from the main features in the item information database to obtain global features and local features;

[0022] Assign different weights to the global features and local features based on the importance of the features;

[0023] Reconstruct the global features and local features based on different weights, and obtain an effective feature image according to the reconstruction result.

[0024] Preferably, in S3, based on the effective feature image, combined with artificial intelligence, the initial classification model is trained to obtain a target classification model, including:

[0025] Based on the matching situation between the effective feature image and the item standard features, establish the feature weights of the effective feature image. Based on the feature weights, train the initial classification model in four ways: from smallest to largest weight, from largest to smallest weight, from smallest to largest difference from the average weight, and from largest to smallest difference from the equal weight, to obtain four corresponding intermediate classification models;

[0026] Based on the convolutional structure and pooling structure of the four intermediate classification models, determine the first influence weight of different convolutional structures on the model performance, and the second influence weight of different pooling results on the model performance;

[0027] Based on the differences between the first influence weights, determine the optimization weights, obtain the optimization parameters of the preset optimization algorithm matching the optimization weights, optimize all the convolutional structures according to the preset optimization algorithm and optimization parameters, and obtain the target convolutional structure according to the optimization results;

[0028] Based on the differences between the second influence weights, determine the algorithm parameters of the adaptive improved pooling algorithm, optimize all the pooling results according to the adaptive improved pooling algorithm and the algorithm parameters, and obtain the target pooling structure according to the optimization results;

[0029] Based on the target convolutional structure and the target pooling structure, establish a target classification model.

[0030] Preferably, in S4, input the collected image of the item to be classified into the target classification model, and determine the category of the item to be classified according to the output result, including:

[0031] Preprocess the collected image of the item to be classified to obtain a target image;

[0032] Input the target image into the target classification model to obtain an output result;

[0033] Determine the category of the item to be classified from the output result.

[0034] Preferably, extract features from the standard image information to obtain main features, including:

[0035] Extract information based on pixel points from the standard image information to obtain the position data and pixel value data of each pixel point;

[0036] Based on the correspondence between the position data and the pixel value data, and combined with the pixel value range, divide the standard image information into multiple image regions, and determine the color numerical feature and color distribution feature of each image region;

[0037] Based on the overall similarity between the color numerical feature and the color distribution feature and the preset saliency feature, assign a first weight to the regions in each image region whose overall similarity meets the preset requirements, and set the weights of other regions to zero to obtain a corresponding first feature image. Based on the channel similarity of the three color channels of the first feature image and the preset saliency feature, assign a second weight to the regions in the first feature image whose channel similarity meets the preset requirements, and set the weights of other regions to zero to obtain a corresponding second feature image;

[0038] Perform feature fusion on the first feature image and the second feature image to obtain a comprehensive feature image, and determine the main features from the comprehensive feature image.

[0039] Preferably, the determination method of the preset significant features is as follows:

[0040] Obtain the common image features in the historical image information of various types;

[0041] Obtain the image features different from the common image features of other historical image information of various types from the common image features as the significant features.

[0042] An item automatic classification system based on artificial intelligence, comprising:

[0043] An information collection module, configured to collect image information of various types of items, extract main features from the image information of various types, and establish an item information database based on the main features;

[0044] A feature processing module, configured to use the RGB grouped attention mechanism to extract and process the main features in the item information database to obtain effective feature images;

[0045] A model establishment module, configured to train an initial classification model based on the effective feature images in combination with artificial intelligence to obtain a target classification model;

[0046] An item recognition module, configured to input the collected image of the item to be classified into the target classification model, and determine the category of the item to be classified according to the output result. Compared with the prior art, the present invention has the following beneficial effects:

[0047] By collecting image information of various types of items, extracting main features from the image information of various types, and establishing an item information database based on the main features, the information features of items in various forms are obtained in advance, providing a rich and comprehensive item information basis for realizing item automatic classification. Using the RGB grouped attention mechanism to extract and process the main features in the item information database to obtain effective feature images, providing accurate and effective feature images for the training of the classification model. Based on the effective feature images, in combination with artificial intelligence, training the initial classification model to obtain a target classification model, inputting the collected image of the item to be classified into the target classification model, and determining the category of the item to be classified according to the output result, quickly and accurately identifying and classifying items, greatly reducing the time and effort of manual classification and improving work efficiency.

[0048] Other features and advantages of the present invention will be set forth in the following description, and in part will be obvious from the description, or may be learned by practice of the present invention. The objectives and other advantages of the present invention may be realized and attained by the structure particularly pointed out in the specification of this application document.

[0049] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0050] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. In the drawings:

[0051] Figure 1 It is a flowchart of an artificial intelligence-based automatic item classification method in an embodiment of the present invention;

[0052] Figure 2 It is a flowchart of collecting various types of image information of items in an embodiment of the present invention;

[0053] Figure 3 It is a structure diagram of an artificial intelligence-based automatic item classification system in an embodiment of the present invention. Detailed Embodiments

[0054] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for explaining and illustrating the present invention, and are not used to limit the present invention.

[0055] Embodiment 1:

[0056] An embodiment of the present invention provides an artificial intelligence-based automatic item classification method, as Figure 1 shown, including:

[0057] S1: Collect various types of image information of items, extract main features from the various types of image information, and establish an item information library based on the main features;

[0058] S2: Use the RGB grouped attention mechanism to extract and process the main features in the item information library to obtain an effective feature image;

[0059] S3: Based on the effective feature image, combine artificial intelligence to train an initial classification model to obtain a target classification model;

[0060] S4: Input the captured image of the item to be classified into the target classification model, and determine the category of the item to be classified according to the output result.

[0061] In this embodiment, collecting various types of image information of an item includes image information under various complex conditions such as various angles, various lighting transformations, complex textures, and occlusions.

[0062] In this embodiment, the main feature is the feature that can represent the uniqueness of the item.

[0063] In this embodiment, the GB grouped attention mechanism is a technique used in image processing, especially in image classification tasks. This mechanism groups the RGB three-channel components and introduces an attention mechanism to enhance the effect of feature extraction.

[0064] The beneficial effects of the above design scheme are as follows: By collecting various types of image information of the item, extracting the main features from the various types of image information, establishing an item information database based on the main features, realizing the advance acquisition of the information features of the item in various forms, providing a rich and comprehensive item information basis for the realization of automatic item classification, using the RGB grouped attention mechanism to extract and process the main features in the item information database to obtain an effective feature image, providing an accurate and effective feature image for the training of the classification model, based on the effective feature image, combined with artificial intelligence, training the initial classification model to obtain a target classification model, inputting the collected image of the item to be classified into the target classification model, and determining the category of the item to be classified according to the output result, quickly and accurately identifying and classifying the item, greatly reducing the time and effort of manual classification and improving work efficiency.

[0065] Embodiment 2:

[0066] Based on Embodiment 1, an embodiment of the present invention provides an artificial intelligence-based automatic item classification method. As Figure 2 shown, in S1, collecting various types of image information of an item includes:

[0067] Collecting various types of image information of the item from historical image collections;

[0068] Using a classification decision tree to classify the various types of image information, and obtaining the image information corresponding to each item according to the classification result.

[0069] The beneficial effects of the above design scheme are as follows: By collecting various types of image information of the item from historical image collections, using a classification decision tree to classify the various types of image information, and obtaining the image information corresponding to each item according to the classification result, providing a rich item information basis for the establishment of the item information database.

[0070] Embodiment 3:

[0071] Based on Example 2, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence. In S1, feature extraction is performed on the various types of image information to obtain main features, and an item information library is established based on the main features, including:

[0072] Preprocess the image information corresponding to each item to obtain standard image information;

[0073] Perform feature extraction on the standard image information to obtain main features;

[0074] Establish an item information library based on the correspondence between the item category of each item and the main features.

[0075] In this embodiment, preprocessing the image information corresponding to each item includes image denoising, filtering, normalization, etc.

[0076] The beneficial effects of the above design are as follows: By preprocessing the image information corresponding to each item to obtain standard image information, performing feature extraction on the standard image information to obtain main features, and establishing an item information library based on the correspondence between the item category of each item and the main features, the information features of various forms of items are obtained in advance, providing a rich and comprehensive item information basis for realizing automatic item classification.

[0077] Example 4:

[0078] Based on Example 1, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence, further including: performing real-time update on the item information library, specifically:

[0079] Obtain the captured images of historical items to be classified in the most recent time period and their corresponding classification result information;

[0080] Match the captured images of the historical items to be classified and their corresponding classification result information with the item information library. If a corresponding item is matched, update the information of the corresponding item based on the captured images of the historical items to be classified and their corresponding classification result information. If no corresponding item is matched, establish new item information in the item information library based on the captured images of the historical items to be classified and their corresponding classification result information.

[0081] The beneficial effects of the above design are as follows: By performing real-time update on the item information library based on the captured images of historical items to be classified in the most recent time period and their corresponding classification result information, the timeliness and comprehensiveness of the item information library are ensured, providing a rich and comprehensive item information basis for realizing automatic item classification.

[0082] Example 5:

[0083] Based on Embodiment 1, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence. In S2, the RGB group attention mechanism is used to extract and process the main features in the item information library to obtain an effective feature image, including:

[0084] The RGB group attention mechanism is used to extract global and local features of the main features in the item information library to obtain global features and local features;

[0085] Based on the importance of the features, different weights are assigned to the global features and local features;

[0086] Based on different weights, the global features and local features are reconstructed, and an effective feature image is obtained according to the reconstruction result.

[0087] In this embodiment, the higher the importance of the feature, the greater the corresponding weight.

[0088] The beneficial effects of the above design are as follows: By using the RGB group attention mechanism to extract global and local features of the main features in the item information library to obtain global features and local features, based on the importance of the features, different weights are assigned to the global features and local features, and based on different weights, the global features and local features are reconstructed, and an effective feature image is obtained according to the reconstruction result, providing a precise and effective feature image for the training of the classification model.

[0089] Embodiment 6:

[0090] Based on Embodiment 1, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence. In S3, based on the effective feature image and combined with artificial intelligence, the initial classification model is trained to obtain a target classification model, including:

[0091] Based on the matching situation between the effective feature image and the standard features of the item, the feature weights of the effective feature image are established. Based on the feature weights, the initial classification model is trained in four ways: from smallest to largest weight, from largest to smallest weight, from smallest to largest difference from the average weight, and from largest to smallest difference from the equal weight, to obtain four corresponding intermediate classification models;

[0092] Based on the convolutional structure and pooling structure of the four intermediate classification models, the first influence weight of different convolutional structures on the model performance and the second influence weight of different pooling results on the model performance are determined;

[0093] Determine the optimization weights based on the differences between the first influence weights, obtain the optimization parameters of a preset optimization algorithm that match the optimization weights, optimize all convolutional structures according to the preset optimization algorithm and the optimization parameters, and obtain the target convolutional structure based on the optimization results;

[0094] Based on the differences between the second influence weights, determine the algorithm parameters of the adaptive improved pooling algorithm, optimize all pooling results according to the adaptive improved pooling algorithm and the algorithm parameters, and obtain the target pooling structure based on the optimization results;

[0095] Based on the target convolutional structure and the target pooling structure, establish the target classification model.

[0096] In this embodiment, the initial classification model is trained in four ways: from smallest to largest weight, from largest to smallest weight, from smallest to largest difference from the average weight, and from largest to smallest difference from the equal-weight weight, to obtain different training effects.

[0097] In this embodiment, feature learning is performed on the convolutional results, and feature selection and extraction are performed on the pooling results.

[0098] In this embodiment, both the preset optimization algorithm and the adaptive improved pooling algorithm are determined in advance, and the specific algorithms vary according to the actual situation.

[0099] In this embodiment, the greater the difference between the first influence weights, the greater the corresponding optimization weights.

[0100] In this embodiment, the optimization parameters of the preset optimization algorithm include, for example, the learning rate, learning factor, etc.

[0101] The beneficial effects of the above design are as follows: By determining the first influence weights of different convolutional structures on the model performance and the second influence weights of different pooling results on the model performance based on the convolutional structures and pooling structures of the four intermediate classification models; determining the optimization weights based on the differences between the first influence weights, obtaining the optimization parameters of a preset optimization algorithm that match the optimization weights, optimizing all convolutional structures according to the preset optimization algorithm and the optimization parameters, and obtaining the target convolutional structure based on the optimization results; determining the algorithm parameters of the adaptive improved pooling algorithm based on the differences between the second influence weights, optimizing all pooling results according to the adaptive improved pooling algorithm and the algorithm parameters, and obtaining the target pooling structure based on the optimization results. Training and optimizing the structure from two aspects of the convolutional structure and the pooling structure to ensure the performance of the obtained target classification model, and providing a model basis for realizing the accurate classification of items to be classified.

[0102] Example 7:

[0103] Based on Embodiment 1, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence. In S4, the captured image of the item to be classified is input into the target classification model, and the category of the item to be classified is determined according to the output result, including:

[0104] Preprocess the captured image of the item to be classified to obtain a target image;

[0105] Input the target image into the target classification model to obtain an output result;

[0106] Determine the category of the item to be classified from the output result.

[0107] The beneficial effect of the above design is that by preprocessing the captured image of the item to be classified to obtain a target image, inputting the target image into the target classification model to obtain an output result, and determining the category of the item to be classified from the output result, the item can be quickly and accurately identified and classified, greatly reducing the time and effort of manual classification and improving work efficiency.

[0108] Embodiment 8:

[0109] Based on Embodiment 3, an embodiment of the present invention provides an automatic item classification method based on artificial intelligence. Feature extraction is performed on the standard image information to obtain main features, including:

[0110] Perform information extraction based on pixel points on the standard image information to obtain the position data and pixel value data of each pixel point;

[0111] Based on the correspondence between the position data and the pixel value data, and in combination with the pixel value range, divide the standard image information into multiple image regions, and determine the color numerical feature and color distribution feature of each image region;

[0112] Based on the overall similarity between the color numerical feature and color distribution feature and the preset saliency feature, assign a first weight to the regions in each image region whose overall similarity meets the preset requirements, and set the weights of other regions to zero to obtain a corresponding first feature image. Based on the channel similarity of the three color channels of the first feature image and the preset saliency feature, assign a second weight to the regions in the first feature image whose channel similarity meets the preset requirements, and set the weights of other regions to zero to obtain a corresponding second feature image;

[0113] Fuse the features of the first feature image and the second feature image to obtain a comprehensive feature image, and determine the main features from the comprehensive feature image.

[0114] In this embodiment, the first feature image is used to determine the overall main features, and the second feature image is used to strengthen the local main features.

[0115] The beneficial effects of the above design scheme are as follows: By based on the correspondence between position data and pixel value data, and combining with the pixel value range, the standard image information is divided into multiple image regions, and the color numerical characteristics and color distribution characteristics of each image region are determined; Based on the overall similarity between the color numerical characteristics and color distribution characteristics and the preset saliency characteristics, regions in each image region whose overall similarity meets the preset requirements are assigned a first weight, and the weights of other regions are zero, obtaining the corresponding first feature image. Based on the channel similarity of the three color channels of the preset saliency characteristics in the first feature image, regions in the first feature image whose channel similarity meets the preset requirements are assigned a second weight, and the weights of other regions are zero, obtaining the corresponding second feature image; The first feature image and the second feature image are subjected to feature fusion to obtain a comprehensive feature image, and the main features are determined from the comprehensive feature image, providing an accurate feature basis for the training and construction of the classification model.

[0116] Example 9:

[0117] Based on Example 8, an embodiment of the present invention provides an artificial intelligence-based automatic item classification method, and the determination method of the preset saliency characteristics is as follows:

[0118] Obtain the common image features in the historical image information of each type;

[0119] Obtain the image features different from the common image features of other historical image information of each type from the common image features as the saliency features.

[0120] The beneficial effects of the above design scheme are as follows: By obtaining the common image features in the image information of each type, and obtaining the image features different from the common image features of other image information of each type from the common image features as the saliency features, it provides a basis for accurately extracting the main features.

[0121] Example 10:

[0122] An embodiment of the present invention provides an artificial intelligence-based automatic item classification system, as Figure 3 shown, including:

[0123] An information collection module, configured to collect the image information of each type of item, extract the main features from the image information of each type, and establish an item information library based on the main features;

[0124] A feature processing module, configured to use the RGB grouped attention mechanism to extract and process the main features in the item information library to obtain an effective feature image;

[0125] A model establishment module, which is used to train an initial classification model based on valid feature images in combination with artificial intelligence to obtain a target classification model;

[0126] An item recognition module, which is used to input the collected image of the item to be classified into the target classification model and determine the category of the item to be classified according to the output result.

[0127] In this embodiment, collecting various types of image information of items includes image information in various complex situations such as various angles, various lighting transformations, complex textures, and occlusions.

[0128] In this embodiment, the main feature is the feature that can represent the uniqueness of the item.

[0129] In this embodiment, the GB grouped attention mechanism is a technique used in image processing, especially in image classification tasks. This mechanism groups the RGB three-channel components and introduces an attention mechanism to enhance the effect of feature extraction.

[0130] The beneficial effects of the above design scheme are as follows: By collecting various types of image information of items, extracting main features from the various types of image information, establishing an item information library based on the main features, realizing the advance acquisition of information features of items in various forms, providing a rich and comprehensive item information basis for realizing automatic item classification, using the RGB grouped attention mechanism to extract and process the main features in the item information library to obtain valid feature images, providing accurate and effective feature images for the training of the classification model, training the initial classification model based on the valid feature images in combination with artificial intelligence to obtain a target classification model, inputting the collected image of the item to be classified into the target classification model, determining the category of the item to be classified according to the output result, quickly and accurately identifying and classifying items, greatly reducing the time and effort of manual classification, and improving work efficiency.

[0131] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of this application document and its equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. An automatic object classification method based on artificial intelligence, characterized in that: include: S1: Collect various types of image information of objects, extract features of the various types of image information to obtain main features, and establish an object information database based on the main features; S2: extracting and processing the main features in the item information database using the RGB grouping attention mechanism to obtain an effective feature image; S3: Based on the effective feature image and combined with artificial intelligence, the initial classification model is trained to obtain the target classification model; S4: Input the collected image of the object to be classified into the target classification model, and determine the category of the object to be classified according to the output result.

2. The method for automatic classification of objects based on artificial intelligence according to claim 1, characterized in that: In S1, various types of image information of objects are collected, including: Collect various types of image information of objects from historical image collection; The various types of image information are classified using a classification decision tree, and the image information corresponding to each object is obtained according to the classification result.

3. The method for automatic classification of objects based on artificial intelligence according to claim 2, characterized in that: In S1, feature extraction is performed on each type of image information to obtain main features, and an item information database is established based on the main features, including: Preprocess the image information corresponding to each object to obtain standard image information; Extracting features from the standard image information to obtain main features; An item information database is established based on the correspondence between the item category and the main features of each item.

4. The method for automatic classification of objects based on artificial intelligence according to claim 1, characterized in that: Also includes: The item information database is updated in real time, specifically: Obtaining historical images of objects to be classified in the most recent period of time and their corresponding classification result information; The captured images of the historical items to be classified and their corresponding classification result information are matched with the item information library. If a corresponding item is matched, the information of the corresponding item is updated based on the captured images of the historical items to be classified and their corresponding classification result information. If no corresponding item is matched, new item information is established in the item information library based on the captured images of the historical items to be classified and their corresponding classification result information.

5. The method for automatic classification of objects based on artificial intelligence according to claim 1, characterized in that: In S2, the main features in the item information database are extracted and processed using the RGB group attention mechanism to obtain a valid feature image, including: Using the RGB grouping attention mechanism, global and local features are extracted from the main features in the item information database to obtain global features and local features; Based on the importance of features, different weights are assigned to the global features and local features; The global features and local features are reconstructed based on different weights, and a valid feature image is obtained according to the reconstruction result.

6. The method for automatic classification of objects based on artificial intelligence according to claim 1, characterized in that: In S3, based on the effective feature image and combined with artificial intelligence, the initial classification model is trained to obtain the target classification model, including: Based on the matching between the effective feature image and the standard feature of the object, the feature weight of the effective feature image is established, and based on the feature weight, the initial classification model is trained in four ways: weight from small to large, weight from large to small, difference from average weight from small to large, and difference from average weight from large to small, to obtain the corresponding four intermediate classification models; Based on the convolution structures and pooling structures of the four intermediate classification models, determining a first influence weight of different convolution structures on model performance, and a second influence weight of different pooling results on model performance; Determine an optimization weight based on the difference between the first influence weights, obtain optimization parameters of a preset optimization algorithm that matches the optimization weight, optimize all convolution structures according to the preset optimization algorithm and the optimization parameters, and obtain a target convolution structure according to the optimization result; Based on the difference between the second influence weights, determining algorithm parameters of the adaptive improved pooling algorithm, optimizing all pooling results according to the adaptive improved pooling algorithm and the algorithm parameters, and obtaining a target pooling structure according to the optimization results; Based on the target convolution structure and the target pooling structure, a target classification model is established.

7. The method for automatic classification of objects based on artificial intelligence according to claim 1, characterized in that: In S4, the collected image of the object to be classified is input into the target classification model, and the category of the object to be classified is determined according to the output result, including: Preprocess the collected images of the objects to be classified to obtain the target images; Inputting the target image into a target classification model to obtain an output result; The category of the object to be classified is determined from the output result.

8. The method for automatic classification of objects based on artificial intelligence according to claim 3, characterized in that: Feature extraction is performed on the standard image information to obtain main features, including: Extracting pixel-based information from the standard image information to obtain position data and pixel value data of each pixel; Based on the correspondence between the position data and the pixel value data and in combination with the pixel value range, the standard image information is divided into a plurality of image regions, and the color value feature and the color distribution feature of each image region are determined; Based on the overall similarity between the color value feature and the color distribution feature and the preset significant feature, a first weight is assigned to the region in each image region whose overall similarity meets the preset requirement, and the weights of other regions are zero, so as to obtain a corresponding first feature image; based on the channel similarity of the three color channels in the first feature image with the preset significant feature, a second weight is assigned to the region in the first feature image whose channel similarity meets the preset requirement, and the weights of other regions are zero, so as to obtain a corresponding second feature image; The first feature image and the second feature image are feature-fused to obtain a comprehensive feature image, and main features are determined from the comprehensive feature image.

9. The method for automatic classification of objects based on artificial intelligence according to claim 8, characterized in that: The preset significant features are determined as follows: Obtain common image features from various types of historical image information; An image feature that is different from the common image features of other types of historical image information is obtained from the common image features as a significant feature.

10. An automatic object classification system based on artificial intelligence, used to implement the steps of the automatic classification method according to any one of claims 1 to 9, characterized in that: include: An information collection module is used to collect various types of image information of objects, extract features of the various types of image information to obtain main features, and establish an object information database based on the main features; A feature processing module, used to extract and process the main features in the item information database using an RGB grouping attention mechanism to obtain an effective feature image; The model building module is used to train the initial classification model based on the effective feature image and combined with artificial intelligence to obtain the target classification model; The object recognition module is used to input the collected images of the objects to be classified into the target classification model and determine the category of the objects to be classified according to the output results.

Images