The Research Of Commodity Image Recognition Based On Deep Learning

High-end buildings and alleys are filled with various retail markets,ranging from high-end supermarkets to department stores and small street convenience stores.The efficient circulation of commodity has improved the quality of life of people and promoted the development of society.Today,with high labor costs,automatic commodity recognition technology is one of the important technologies for ensuring the efficient circulation of commodity.The most mature commodity recognition technology at this stage is bar code recognition,but it is generally necessary to manually align the bar code position to scan code recognition,and there is a low degree of automation.Moreover,the barcode itself has problems such as being easily deformed and easily damaged,which affects the recognition accuracy.Radio frequency recognition technology(RFID)is an automatic recognition technology that solves the problems of barcode technology.When RFID is used to identify commodities,the cost of the label is too high,such as the cost of the label itself,the labor cost of labeling,and the time cost of labeling.On average,a single item of merchandise occupies a certain proportion of the price of a certain commodity,especially if the profit of a single item of retail merchandise is extremely low,and the cost of the label further reduces the profit of the merchant.In recent years,image processing technology and deep learning technology have been rapidly developed,and there are no signs such as barcodes,labels,and the like,and a large number of commodity images are used to learn the characteristics of the commodity packaging itself,so that the computer has the ability to automatically recognize commodity images.This recognition technology has the advantages of high degree of automation,low cost,high accuracy,and has very important research significance in the coming day of “unmanned retail”.The main contents of this paper include: Firstly,the background and significance of the research of this paper is clear,the current commodity recognition technology is learned and learned,combined with the advantages of deep learning and image processing technology,a commodity image recognition method based on deep learning and image processing is proposed..The image features of the commodity packaging were studied,and several commonly used visual features were compared and analyzed.Several basic models of neural networks are introduced.Convolution effects of different kernels on convolutional images in convolutional neural networks are compared,and the pooling process is introduced.The principle and training process of deep learning technology are summarized,and the main models and frameworks are analyzed and selected.Then,combining the appearance characteristics of various commodities,designing targeted commodity shooting schemes and experimental platforms for image acquisition,commodity image preprocessing and image expansion technologies are studied to generate commodity image data sets.Finally,according to the self-built commodity image database,the model of the commodity recognition system is trained on the network.The accuracy of the cross validation set reaches 81.24%.And a brief description of the practical application of the commodi ty recognition system,compared and analyzed the advantages and disadvantages of the two application model methods in the commercial phase.