Open commodity visual identification method and system
By building a product feature library containing different shooting angles and placement methods, combining deep learning and traditional image processing, the problem of poor product visual recognition effect in the prior art is solved, and high robustness and low-cost product recognition are achieved.
Patent Information
- Application Number
- CN202411980933.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-07-11
AI Technical Summary
The existing open visual recognition technology for goods is not effective in handling the random placement of goods and reducing deployment costs.
By obtaining the feature library of different shooting angles and placement methods of products in the base library, conducting statistical analysis, obtaining the characteristics of the products to be identified, and category recognition is carried out based on the weight statistical information, and using a combination of deep learning and traditional image processing to improve the adaptability and accuracy of the recognition system.
It significantly improves the adaptability and accuracy of the product identification system, reduces deployment costs, and achieves high robustness recognition of different shooting angles and placement methods.
Smart Images
Figure CN120298740A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of visual recognition, and particularly to an open commodity visual recognition method and system. Background Art
[0002] The open commodity visual recognition technology refers to the recognition technology that can detect all commodities. The application of the open commodity visual recognition technology is relatively extensive, such as various forms of self-service cash register systems, inventory monitoring systems for goods in supermarket shelves, and so on. And this field involves a wide variety of commodity types, and there are thousands of differences in the appearance between commodities, which increases the difficulty of applying the open commodity visual recognition technology. This technology application field has also received extensive attention, and many technical solutions have been proposed: For a specific supermarket, collect all commodity image data, train a specific commodity target detection model, and train a specific commodity classification model. Connect the two models to implement commodity recognition applications. The target detection finds all commodity targets in the field of view, and the commodity classification realizes the recognition of all commodities in the field of view; With the improvement of the measurement model effect, on the basis of the foregoing solution, replace the commodity classification model with a measurement model. The advantage of this is that the measurement model can be a general commodity measurement model, and there is no need to train a specific commodity classification model for different scenarios as in the foregoing solution, which greatly reduces the application cost.
[0003] However, the recognition effects of most current solutions are not good, and they cannot handle the random placement of commodities well, and do not well reduce the deployment cost of specific application scenarios. Summary of the Invention
[0004] The main purpose of the present invention is to solve the technical problem that the recognition effects of most existing solutions are not good. An open commodity visual recognition method includes the following steps: Obtain a commodity base feature library of the base commodities with different shooting angles and different placement methods; perform statistics on the commodity base feature library to obtain the weight statistics information of different shooting angles and different placement methods of each category of commodities; obtain the features of the commodity to be recognized ; complete the category recognition of the commodity to be recognized according to the weight statistics information of different shooting angles and different placement methods.
[0005] As a preferred technical solution, obtaining a commodity base feature library of the base commodities with different shooting angles and different placement methods includes: An open commodity target detection model is adopted to extract the commodity pictures to be recognized with different shooting angles and different placement methods; the commodity pictures to be recognized are corrected to obtain the corrected commodity pictures; high discriminative features of the target are extracted from the corrected commodity pictures to obtain the commodity base feature library of the bottom library commodities with different shooting angles and different placement methods.
[0006] As a preferred technical solution, adopting an open commodity target detection model to extract the commodity pictures to be recognized with different shooting angles and different placement methods includes: According to the commodity type and detection target, obtain the commodity images that meet the conditions; select the target detection annotation tool to manually annotate the obtained commodity images, including drawing bounding boxes, annotating commodity categories, shooting angles, and placement methods; preprocess the commodity images, and divide the preprocessed data into a training set, a validation set, and a test set; select a suitable convolutional neural network and a network structure based on Transformer as the model basis; configure the training parameters: set appropriate training parameters such as learning rate, batch size, and number of training epochs; use the training set to train the model, monitor the performance and adjust the parameters through the validation set to obtain a trained open commodity target detection model; adopt the trained open commodity target detection model to extract the commodity pictures to be recognized with different shooting angles and different placement methods.
[0007] As a preferred technical solution, correcting the commodity pictures to be recognized to obtain the corrected commodity pictures includes: Input the commodity pictures to be recognized into the target segmentation model, and run the target segmentation model for target segmentation to generate a pixel-level segmentation mask. Use the segmentation mask obtained in the previous step to cut out the commodity target from the original picture; based on the contour of the cut-out result, find the leftmost, rightmost, topmost, and bottommost pixel points of the contour to determine the coordinates of the minimum bounding rectangle; according to the coordinates of the determined minimum bounding rectangle, rotate the cut-out commodity target so that it faces the observer directly; check whether the height of the rotation result is greater than the width, if not, rotate it by 90 degrees; use the rotated and corrected commodity target as the corrected commodity picture.
[0008] As a preferred technical solution, extracting high discriminative features of the target from the corrected commodity pictures to obtain the commodity base feature library of the bottom library commodities with different shooting angles and different placement methods includes: Select a suitable feature extraction model according to the characteristics of the commodity pictures and the task requirements; through the feature extraction model, extract high discriminative features from commodity pictures with different shooting angles and different placement methods; store the extracted high discriminative features together with the corresponding commodity information into the commodity base feature library.
[0009] As a preferred technical solution, the commodity bottom library feature library is statistically analyzed to obtain the weight statistical information of different shooting angles and different placement methods of commodities in each category, including: Extract the bottom library features of each shooting angle and each placement method of the bottom library commodities ; Calculate the average feature of all shooting angles under each placement method of the bottom library commodities ; Calculate the weight of the features at each shooting angle under each placement method of the bottom library commodities ; The weight is obtained by calculating the similarity between the average feature and the bottom library feature .
[0010] As a preferred technical solution, according to the weight statistical information of different shooting angles and different placement methods, the category identification of the commodity to be identified is completed, including: Calculate the similarity between the commodity to be identified and different placement methods of the bottom library commodities : Among them, is the similarity between the feature and the bottom library feature , i represents the placement method, j represents the shooting angle, and n represents the number of pictures of the shooting angle under the placement method; Calculate the similarity between the commodity to be identified and the bottom library commodities : Among all the bottom library commodities, the category of the bottom library commodity with the largest value is determined as the category of the commodity to be identified.
[0011] The second aspect of the present invention provides an open commodity visual recognition device, including: A feature library construction unit for obtaining a commodity bottom library feature library of different shooting angles and different placement methods of bottom library commodities; a feature library statistical unit for statistically analyzing the commodity bottom library feature library to obtain the weight statistical information of different shooting angles and different placement methods of commodities in each category; a feature acquisition unit for the commodity to be identified for obtaining the features of the commodity to be identified ; A category identification unit for completing the category identification of the commodity to be identified according to the weight statistical information of different shooting angles and different placement methods.
[0012] A third aspect of the present invention provides an electronic device, including: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected by a line; the at least one processor invokes the instructions in the memory to enable the electronic device to execute the above-mentioned open commodity visual recognition method.
[0013] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, and when it runs on a computer, it enables the computer to execute the above-mentioned open commodity visual recognition method.
[0014] The present invention has the following beneficial effects: A novel and perceptive image quality assessment method based on the combination of deep learning and traditional image processing proposed in this patent belongs to the no-reference image quality assessment method. It mainly introduces less artificial supervision and at the same time achieves better image quality assessment results.
[0015] Traditional commodity recognition schemes often only consider the front or fixed-angle pictures of commodities, ignoring the diversity of commodities under different shooting angles and placement methods. The solution of the present invention significantly improves the adaptability and accuracy of the recognition system by obtaining the features of bottom-library commodities under different shooting angles and placement methods.
[0016] By statistically analyzing the bottom-library feature library of commodities, the weight statistical information of each category of commodities under different shooting angles and placement methods is obtained. Through this method, the system can more accurately understand the visual feature distribution of different categories of commodities, providing strong data support for subsequent category recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 The first flowchart of the open commodity visual recognition method provided by the embodiment of the present invention; Figure 2 The open commodity recognition result diagram; Figure 3 The open commodity recognition correction diagram; Figure 4 The structural schematic diagram of the open commodity visual recognition device provided by the embodiment of the present invention; Figure 5 The structural schematic diagram of the electronic device provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0018] The embodiment of the present invention provides an open commodity visual recognition method and system. The method includes: Obtain the product base feature library of the base library products with different shooting angles and different placement methods; perform statistics on the product base feature library to obtain the weight statistics information of different shooting angles and different placement methods for each category of products; obtain the features of the product to be recognized ; According to the weight statistics information of different shooting angles and different placement methods, complete the category recognition of the product to be recognized. This patent proposes an open commodity visual recognition method with low cost and high robustness.
[0019] The terms "first", "second", "third", "fourth", etc. (if any) in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order different from that shown or described herein. In addition, the term "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.
[0020] Open commodity visual recognition It means that this recognition method can recognize all commodities.
[0021] For ease of understanding, the specific process of the embodiment of the present invention will be described below. Please refer to Figure 1 In the first embodiment of the open commodity visual recognition method in the embodiment of the present invention, it includes: 101. Obtain the product base feature library of the base library products with different shooting angles and different placement methods; Step 1: Obtain the product base feature library of the base library products with different shooting angles and different placement methods In this step, a feature library containing various products under different shooting angles and placement methods is constructed.
[0022] This step collects a large number of product images, ensuring that these images cover different categories, shooting angles, and placement methods. Use computer vision techniques (such as deep learning models) to extract features from each image. These features can be low-level features such as the texture, shape, and color of the image, or high-level features learned through deep learning models. Store the extracted features together with the corresponding product information (such as category, shooting angle, placement method, etc.) to form a feature library of the product base. Use an open product object detection model to extract product images to be recognized at different shooting angles and different placement methods; correct the product images to be recognized to obtain corrected product images; extract highly discriminative features of the target from the corrected product images to obtain a feature library of the product base for different shooting angles and different placement methods of the products in the base library.
[0023] More specifically, it includes the following steps: According to the product type and detection target, obtain qualified product images; select an object detection annotation tool to manually annotate the obtained product images, including drawing bounding boxes, annotating product categories, shooting angles, and placement methods; preprocess the product images, and divide the preprocessed data into a training set, a validation set, and a test set; select a suitable convolutional neural network and a network structure based on Transformer as the model foundation; configure training parameters: set appropriate training parameters, such as learning rate, batch size, and number of training epochs; use the training set to train the model, monitor the performance through the validation set and adjust the parameters to obtain a trained open product object detection model; use the trained open product object detection model to extract product images to be recognized at different shooting angles and different placement methods. Figure 2 。
[0024] Input the product image to be recognized into the target segmentation model, and run the target segmentation model for target segmentation to generate a pixel-level segmentation mask. Use the segmentation mask obtained in the previous step to cut out the product target from the original image; based on the contour of the cut-out result, find the leftmost, rightmost, topmost, and bottommost pixel points of the contour to determine the coordinates of the minimum bounding rectangle; according to the determined coordinates of the minimum bounding rectangle, rotate the cut-out product target so that it faces the observer; check whether the height of the rotation result is greater than the width, if not, rotate it by 90 degrees; use the rotated and corrected product target as the corrected product image. Figure 3 。
[0025] Select a suitable feature extraction model according to the characteristics of the product image and the task requirements; through the feature extraction model, extract highly discriminative features from product images at different shooting angles and different placement methods; store the extracted highly discriminative features together with the corresponding product information in the feature library of the product base.
[0026] There are usually multiple ways to arrange goods. For example Figure 2 In the case of crispy rice fragrance with a regular paper box, which has six sides, there are six ways to arrange it. In this patent, it is required that the bottom library pictures cover the appearances of all arrangement ways of the target goods. Another example is common bagged goods (such as potato chips, etc.), which usually have two arrangement ways (front and back). At the same time, in order to improve the accuracy of goods recognition, pictures of multiple shooting angles for each side will be collected. Suppose a good has m sides, and n shooting angles are provided for each side, then a total of m*n bottom libraries are provided for this good (this is a special case here, and the number of shooting angles for each side can be different, as long as n>=1 is ensured).
[0027] Feature extraction of bottom library pictures: Here, for each bottom library picture, the steps introduced in the previous process will be carried out: [Open commodity target detection]->[Commodity target correction]->[Commodity target feature extraction], and finally a commodity bottom library feature library is obtained.
[0028] 102. Statistically analyze the commodity bottom library feature library to obtain the weight statistical information of different shooting angles and different arrangement ways of each category of goods; The purpose of this step is to analyze the influence of different shooting angles and arrangement ways on the recognition of commodity categories and determine their weights.
[0029] Data cleaning and preprocessing: Clean the feature library to remove duplicate or invalid data and perform necessary preprocessing. Conduct statistical analysis on the data in the feature library, calculate the appearance frequency or proportion of each category of goods at different shooting angles and arrangement ways. According to the results of the statistical analysis, determine the weights of different shooting angles and arrangement ways. The weights can be determined according to factors such as appearance frequency and recognition accuracy rate.
[0030] As a preferred implementation method, extract the bottom library features of each shooting angle under each arrangement way of the bottom library goods ; calculate the average feature of all shooting angles under each arrangement way of the bottom library goods ; calculate the weight of the feature at each shooting angle under each arrangement way of the bottom library goods ; the said weight is obtained by calculating the similarity between the average feature and the bottom library feature .
[0031] In this embodiment, the specific steps are as follows: Taking a certain good as an example, assume it has m arrangement ways, and n-angle pictures are taken under each arrangement way. After the feature extraction of the bottom library pictures, this good has a total of m*n bottom library features, that is : Among them, 。
[0032] The steps for bottom library feature statistics are as follows: ① Calculate the average feature of , that is, calculate: ② Calculate the weight of each feature in Among them, means calculating the similarity between and . means the reference degree (weight) of the bottom library features of each shooting angle under the i-th placement method of the commodity.
[0033] According to this method, add the bottom library features and statistical information of each category of commodities. The bottom library features of the c-th commodity are .
[0034] 103. Obtain the features of the commodity to be recognized ; The purpose of this step is to extract the features of the image of the commodity to be recognized for category recognition. Necessary preprocessing is performed on the image of the commodity to be recognized, such as scaling, cropping, denoising, etc. The features of the image of the commodity to be recognized are extracted using the same method as when constructing the feature library. It can be processed in the same way as the construction method of the aforementioned commodity bottom library feature library.
[0035] 104. Complete the category recognition of the commodity to be recognized according to the weight statistical information of different shooting angles and different placement methods.
[0036] This step uses the weight statistical information and the features of the commodity to be recognized for category recognition. The features of the commodity to be recognized are matched with the features in the commodity bottom library feature library to find the most similar commodity or feature set. According to the weight statistical information, the matching results under different shooting angles and placement methods are weighted and adjusted. According to the weighted and adjusted matching results, the category of the commodity to be recognized is determined.
[0037] As a preferred technical solution, completing the category recognition of the commodity to be recognized according to the weight statistical information of different shooting angles and different placement methods includes: Calculate the similarity between the commodity to be recognized and the bottom library commodities in different placement methods : Among them, as a feature and the feature of the bottom database The similarity, where i represents the placement method, j represents the shooting angle, and n represents the number of pictures of the shooting angle under the placement method; Calculate the similarity between the product to be recognized and the product in the bottom database : Among all the products in the bottom database, Determine the category of the product in the bottom database with the largest value as the category of the product to be recognized.
[0038] Calculation of similarity: Euclidean Distance: The Euclidean distance is one of the most commonly used distance metrics and is used to calculate the straight-line distance between two points. In the feature space, you can calculate the Euclidean distance between two feature vectors as a measure of their dissimilarity. The smaller the distance, the higher the similarity.
[0039] Formula: (d(x, y) = \sqrt{\sum_{i=1}^{n} (x_i - y_i)^2}) where (x) and (y) are two feature vectors and (n) is the number of features.
[0040] Cosine Similarity: Cosine similarity measures the similarity between two vectors by calculating the cosine value of the angle between them. This method focuses more on the similarity in direction rather than the absolute magnitude.
[0041] Formula: (\text{similarity} = \cos(\theta) = \frac{\mathbf{A} \cdot \mathbf{B}}{|\mathbf{A}| |\mathbf{B}|}) where (\mathbf{A}) and (\mathbf{B}) are two feature vectors, (\cdot) represents the dot product, and (|\cdot|) represents the magnitude of the vector.
[0042] Jaccard Similarity: For set-type data, Jaccard similarity is a commonly used metric. It calculates the ratio of the size of the intersection of two sets to the size of the union.
[0043] Formula: \(J(A, B)=\frac{|A\cap B|}{|A\cup B|}\) where \(A\) and \(B\) are two sets.
[0044] Specifically: For the base library features of multiple shooting perspectives under each placement method of the commodity and the weights of the features under each shooting angle . The comparison method between the target features of the commodity and the base library features of the commodity is as follows: (Taking the comparison with a single commodity base library as an example) Suppose the features of the commodity to be recognized are Calculate The similarity with each category of commodities If it is most similar to a certain category of commodities, then it can be known that the recognized commodity category of the commodity is consistent with this commodity category (i.e., the commodity recognition result).
[0045] Specifically, first calculate The similarity with each placement method of the commodity: ; Finally, The similarity with a single commodity: .
[0046] Compare with each commodity in the base library in turn to obtain multiple , and finally The commodity category corresponding to the maximum value is The corresponding recognition category.
[0047] In the real world, the shooting angles and placement methods of commodities are often diverse due to factors such as shooting conditions, photographer habits, and display requirements. This diversity is not only reflected in the appearance of the commodities but also directly affects the accuracy of feature extraction and recognition. Therefore, considering the weight statistical information of different shooting angles and placement methods can more realistically reflect the diversity of commodities in actual scenarios, thereby improving the accuracy of category recognition.
[0048] By statistically analyzing the occurrence frequency or proportion of different shooting angles and placement methods in the commodity base library, their importance in category recognition can be determined. Higher-weight shooting angles and placement methods mean that they have a greater contribution in the recognition process, so more attention should be paid to them. Using this weight information, different features can be weighted during the recognition process, thereby improving the recognition accuracy.
[0049] The above describes the open commodity visual recognition method in the embodiments of the present invention. Next, the open commodity visual recognition device in the embodiments of the present invention will be described. Please refer to Figure 4 , the first embodiment of the open commodity visual recognition device in the embodiments of the present invention includes: A feature library construction unit, which is used to obtain the commodity base feature library of the base library commodities at different shooting angles and different placement methods; a feature library statistics unit, which is used to statistically analyze the commodity base feature library to obtain the weight statistics information of different shooting angles and different placement methods of each category of commodities; a to-be-recognized commodity feature acquisition unit, which is used to acquire the features of the to-be-recognized commodity ; a category recognition unit, which is used to complete the category recognition of the to-be-recognized commodity according to the weight statistics information of different shooting angles and different placement methods.
[0050] Figure 5 FIG. is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 500 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 510 (for example, one or more processors) and a memory 520, and one or more storage media 530 for storing application programs 533 or data 532 (for example, one or more mass storage devices). Among them, the memory 520 and the storage media 530 may be transient storage or persistent storage. The program stored in the storage media 530 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the electronic device 500. Further, the processor 510 may be configured to communicate with the storage media 530 and execute a series of instruction operations in the storage media 530 on the electronic device 500.
[0051] The electronic device 500 may further include one or more power supplies 540, one or more wired or wireless network interfaces 550, one or more input / output interfaces 550, and / or one or more operating systems 531, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art can understand that Figure 5 the shown structure of the electronic device does not constitute a limitation on the electronic device, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.
[0052] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are run on a computer, the computer is caused to execute the steps of the open commodity visual recognition method.
[0053] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device, or unit can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.
[0054] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention. The foregoing storage medium includes: USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.
[0055] As described above, the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or equivalently replace some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of the present invention.
Claims
1. An open commodity visual recognition method, characterized in that, Including the following steps: Obtain the product base feature library of the base library products with different shooting angles and different placement methods; Statistically analyze the product base feature library to obtain the weight statistical information of different shooting angles and different placement methods for each category of products; Obtain the features of the product to be recognized ; Complete the category identification of the product to be recognized according to the weight statistical information of different shooting angles and different placement methods.
2. The open commodity visual recognition method according to claim 1, characterized in that The obtaining of the product base feature library of the base library products with different shooting angles and different placement methods includes: Adopt an open product target detection model to extract the product pictures to be recognized with different shooting angles and different placement methods; Rectify the product pictures to be recognized to obtain the rectified product pictures; Extract the highly discriminative features of the target from the rectified product pictures to obtain the product base feature library of the base library products with different shooting angles and different placement methods.
3. The open commodity visual recognition method according to claim 2, characterized in that The adopting of an open product target detection model to extract the product pictures to be recognized with different shooting angles and different placement methods includes: According to the product type and detection target, obtain the qualified product images; Select a target detection annotation tool to manually annotate the obtained product images, including drawing bounding boxes, annotating product categories, shooting angles, and placement methods; Preprocess the product images and divide the preprocessed data into a training set, a validation set, and a test set; Select a suitable convolutional neural network and a Transformer-based network structure as the model basis; Configure the training parameters: set appropriate training parameters, such as learning rate, batch size, and number of training epochs; Use the training set to train the model, monitor the performance through the validation set and adjust the parameters to obtain a trained open product target detection model; Adopt the trained open product target detection model to extract the product pictures to be recognized with different shooting angles and different placement methods.
4. The open commodity visual recognition method according to claim 2, wherein The rectifying of the product pictures to be recognized to obtain the rectified product pictures includes: Input the product pictures to be recognized into the target segmentation model and run the target segmentation model for target segmentation to generate a pixel-level segmentation mask; Use the segmentation mask obtained in the previous step to cut out the product target from the original picture; Based on the contour of the cut-out result, find the leftmost, rightmost, topmost, and bottommost pixel points of the contour to determine the coordinates of the minimum bounding rectangle; According to the coordinates of the determined minimum bounding rectangle, rotate the cut-out product target so that it faces the observer; Check whether the height of the rotation result is greater than the width. If not, perform a 90-degree rotation; Use the rotation-rectified product target as the rectified product picture.
5. The open commodity visual recognition method according to claim 2, characterized in that, The extracting of the highly discriminative features of the target from the rectified product pictures to obtain the product base feature library of the base library products with different shooting angles and different placement methods includes: Select a suitable feature extraction model according to the characteristics of the product pictures and the task requirements; Extract highly discriminative features from product pictures with different shooting angles and different placement methods through the feature extraction model; Store the extracted highly discriminative features together with the corresponding product information into the product base feature library.
6. The open commodity visual recognition method according to claim 1, wherein The statistical analysis of the product base feature library to obtain the weight statistical information of different shooting angles and different placement methods for each category of products includes: Extract the base library features for each placement method and each shooting angle of the base library products ; Calculate the average features of all shooting angles for each placement method of the products in the bottom library ; Calculate the weights of features for each placement method and each shooting angle of the products in the bottom library ; The weights are obtained by calculating the similarity between the average feature and the bottom library feature .
7. A visual recognition method for open commodities according to claim 1, characterized in that Performing category recognition of a commodity to be recognized based on the weight statistical information of different shooting angles and different placement manners, including: Calculate the similarity between the commodity to be recognized and the bottom library commodities in different placement manners : Among them, is the feature and the similarity of the bottom database feature where i represents the placement method, j represents the shooting angle, and n represents the number of pictures of the shooting angle under the placement method; Calculate the similarity between the product to be recognized and the products in the base library : Among all the products in the bottom database, The category of the product in the bottom database with the largest value is determined as the category of the product to be recognized.
8. An open commodity visual recognition device, characterized in that, The device includes: A feature library construction unit configured to obtain a commodity base feature library of the base library commodities with different shooting angles and different placement manners; A feature library statistics unit configured to perform statistics on the commodity base feature library to obtain the weight statistical information of different shooting angles and different placement manners of each category of commodities; Unit for obtaining characteristics of commodity to be recognized, which is used to obtain the characteristics of the commodity to be recognized ; A category recognition unit configured to complete the category recognition of the commodity to be recognized according to the weight statistical information of different shooting angles and different placement manners.
9. An electronic device, the electronic device includes a memory and at least one processor, and instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the electronic device to execute each step of the open commodity visual recognition method according to any one of claims 1-7.
10. A computer-readable storage medium, on which instructions are stored, characterized in that, When the instructions are executed by the processor, each step of the open commodity visual recognition method according to any one of claims 1-7 is implemented.