Commodity Recognition System Of Intelligent Vending Machine Based On Deep Learning

The unmanned retail industry has developed rapidly,bringing convenience to people's lives.However,there are some shortcomings in the vending equipment in the current industry.Traditional vending machines are cumbersome to purchase goods,and the lower outlet setting makes it inconvenient to take goods.Although the container based on RFID(Radio Frequency Identification)solves the problem of non-contact identification,it increases the cost of RFID tags and the labor cost of labeling.In addition,the vision-based containers in the industry need to be divided into goods lanes,and there are problems such as fewer types of goods sold,low container space utilization,large amount of data calculation,low recognition rate,serious misidentification and missed recognition.In view of the shortcomings of the existing unmanned retail equipment,this paper designs a smart container commodity identification system based on deep learning,which is used to simplify the purchase steps and reduce the cost of commodity identification.This system mainly processes the shopping video taken by the container.In order to filter out the interference of the external motion of the container on the motion detection and avoid the influence of the image in the container on the classification of the goods,the ROI(Region of Interest)is selected in the middle of the video frame.The selection of the ROI reduces the data processing while avoiding interference the amount.Commodity recognition needs to capture the motion and get the picture of taking the commodity,so the frame difference method and image morphology processing are used to process the video ROI.After obtaining the movement area,in order to judge the customer's product taking action,the coordinates of the movement area will be statistically analyzed,and the Single Shot Multi Box Detector(SSD)model will be used to assist the judgment of manual target detection.If you want to know what the customer is taking,use the classification network Inception V3 to classify the pictures to be classified by motion detection.In order to ensure the accuracy of the classification results of commodities,the classification results in a process are counted,and then the types of commodities taken by customers in the process are obtained.Finally,the product classification results and action judgments are integrated to give the products recognized by the system.Through the test of the entire system,the system can achieve a better recognition effect for the recognition of limited scenes.The accuracy rate of shopping video product recognition with a single product reached 74%,and the overall recognition rate of shopping video product recognition reached 64%.Combined with the hardware equipment of the merchant,it can give recognition abnormalities to unrecognized products,which makes the product recognition system more robust.