Application Research Of Commodity Recognition Method Based On Deep Learning

With the rapid development of AI,Cloud Computing,Computer Vision and other technologies,all walks of life are using new technologies to improve production efficiency and service quality.Commodity consumption is a scene closely related to people's lives.Commodity identification,as a key part of it,is related to consumers 'consumption experience and merchants' service quality.Existing self-service checkout equipment uses barcode as the main identification method,which has the problem of low identification efficiency;and manual management of commodity shelves may cause untimely replenishment of commodities.Therefore,research on commodity recognition methods based on deep learning has high application value and significance.This thesis focuses on the application of the product identification method research work as follows:1.In the commodity settlement scenario,currently the self-service settlement equipment mainly uses barcode as the main commodity identification method.For ordinary consumers,each barcode needs to be aligned with the scanning area of the device,which undoubtedly increases the complexity of the settlement and reduces the convenience.This paper studies the image classification method based on deep learning as a commodity recognition method,and realizes the commodity recognition function by directly classifying the outer packaging of the commodity.First of all,this article collects and produces a commodity image classification data set.Aiming at the data set of this article,the current mainstream Res Net50 and Res Net101 networks are respectively transferred to the learning training.Finally,by comparing the network performance under the test data,Res Net50 achieves 97.4% precision and 97.2% recall;Res Net101 achieves 97.8% precision and 97.7% recall.2.In the shelf management scenario,current convenience equipment such as automatic vending containers mainly rely on manual management,which has a low degree of automation and is prone to the problem of untimely shelf replenishment.Aiming at this problem,this paper studies the target detection method based on deep learning as a shelf monitoring method,and realizes the shelf status reminding function by monitoring the types and quantities of goods in the target area.First of all,this article made a commodity target detection data set and annotated the image data.Aiming at the dataset of this article,this article trains the current mainstream Two-Stage target detection algorithm Faster R-CNN and compares the performance with the Mobilenetv2-SSDlite lightweight model.Finally,under the test data,Faster R-CNN achieves 0.9787 m AP and Mobilenetv2-SSDlite achieves 0.9510 m AP,and the Faster R-CNN model performs better than Mobilenetv2-SSDlite in more complex tasks.3.In this paper,based on the trained Res Net50 commodity image classification model and Faster R-CNN commodity target detection model,we design the commodity retail settlement system and the intelligent shelf reminder system simulation program respectively.Among them,we add My SQL database to the retail settlement system for the recording and management of commodity settlement pipelines;we also add the voice synthesis technology to the shelf reminder system to achieve the voice prompt function.Finally,the simulation programs is tested and the expected function is achieved.