Research On Commodity Detection And Classification Algorithm Based On Online Update

The retail industry has always been an important part of my country's economic development.However,with the development of e-commerce in recent years,online transactions have begun to flourish,and offline retail transactions have gradually declined.Searching the information found that a large part of the reason why people reduce offline shopping is because if there are a lot of people in line at the checkout,it will take a long time at the checkout,resulting in a poor shopping experience.At present,the checkout methods adopted by most supermarkets are traditional barcode recognition technology and radio frequency technology.The barcode technology requires the salesperson to hold the scanner to scan the barcode on the product,and the radio frequency technology uses the sensor to sense the electronic label on the product.Both of these methods still require manual operation and have not been fully automated.As the image recognition technology based on deep learning has been greatly improved compared with before,it can meet the needs of various visual application scenarios.In order to solve the problem of slow checkout and longer queue time when shopping in the supermarket.A self-service checkout system based on deep learning came into being.Customers only need to place the purchased goods on the checkout counter,and the system will automatically scan the barcode to identify the product category,and finally generate the price to complete the checkout.However,there is a drawback of this system.If the supermarket introduces new products or replaces new packages with the same products,the previously trained model is no longer applicable to the current scenario.If only the new product training model is used,the model will appear.Catastrophic forgetting;if you use all the goods to retrain,it will waste resources and time.Moreover,at the time of checkout,the placement of the products purchased by the customer may cause the products to block each other,which will affect the detection and recognition accuracy of the system.The task of this article is to solve how to avoid the repeated training of the model caused by the introduction of new products and reduce the interference of the product occlusion so as to detect the products and then correctly classify them.Using deep learning technology to design an online update that only needs to be trained once Product detection and classification algorithms.Finally,we verified the effectiveness of our proposed algorithm on the D2 S and RPC datasets,and finally combined these two models.Experiments found that the model performs best when the comparative learning loss function is used.