Refrigerator commodity image recognition training method, recognition method and share calculation method
A product image and training method technology, applied in the field of image recognition, can solve problems such as the inability to quickly calculate the market share of a specific type of product, and the inability to identify different types of products
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0029] Such as figure 1 As shown, the present invention provides a refrigerator product image recognition training method for recognizing cold drink products in the freezer, including the following steps. S10: Obtain a freezer product image containing multiple cold drink products, input the first neural network model, the first neural network model labels clear cold drink products, and obtains multiple labeled polygon borders (polygon in image processing), polygon The border encloses the cold drink product into a closed polygon. S20: Analyze the polygonal frame to obtain the mapping of cold drink products on multiple feature layers, the initial training image and the corresponding mask label; S30: Input the initial training image into the initial neural network model to obtain the cold drink product in the same format as the mask label Image data; S40: Substituting the cold drink commodity image data and mask label data into the loss function calculation, and returning the lo...
Embodiment 2
[0038] A method for image recognition of refrigerator commodities, such as figure 2 As shown, the recognition method is an improved model with instance segmentation function based on yolov5, and the collected images of freezer products containing multiple cold drink products are trained by the training method provided by the present invention. The specific process is as follows. S100: Input the collected freezer product images, perform image size adjustment and standardization processing, and send them to the backbone convolutional neural network to extract features to obtain the first feature layer; S200: Obtain the second feature layer of different sizes through the first feature layer Feature layer, third feature layer; S300: Use each pixel in the first feature layer, the second feature layer, and the third feature layer as a grid, and generate relative coordinates in the x-coordinate direction and y-coordinate direction for each grid. The first offset of the upper left co...
Embodiment 3
[0041] A method for calculating the share of refrigerator commodities, such as image 3As shown, it includes the following steps: S1000: Collecting images of the refrigerators containing commodities, obtaining images of the commodities in the refrigerators, and judging whether the image quality of the commodities in the refrigerators is qualified. Existing technology can be used for judging whether the quality is qualified, or manual judgment can be used. The main criteria The image needs to include most of the cold drink products in the freezer and be clear, and can accurately distinguish most of the cold drink products in the freezer. If qualified, go to step S2000, otherwise go to step S7000. S2000: Adjust the size of the freezer product image to 608*608 pixels, and perform data enhancement processing to obtain a preprocessed image. After the preprocessing, the freezer product image is processed in the second neural network model. S3000: Input the preprocessing image into ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


