Identification Of Beverage Product In Smart Containers Based On Deep Learning

Based on intelligent detecting a beverage goods container inside the vending machine,and provides a solution based on depth of the neural network vision,purpose is to help enterprises with more security and flexible at the same time at a lower cost to the smart container vending machine for quick deployment and beverage type quick update iteration,at the same time in such aspects as offline self-help shopping experience for consumers is extremely vital significance.The identification of beverage commodities designed in this paper is mainly divided into three parts: data set collection,labeling and evaluation,detection target,and classification recognition.The specific content is divided into the following three:(1)In order to obtain the entire container field of view,this article uses fisheye lens to collect data and correct it,and then evaluate the distribution of the data set between categories and categories.(2)For the data set of the detection model,this paper adopts the detection algorithm of SSD as the beverage product,through the analysis of the algorithm principle and the characteristics of the beverage product,and has made corresponding improvements to adapt to the beverage data set,Trained a network model specifically suitable for beverage detection,and obtained the ideal detection effect.(3)For the data set of the classification model,this paper refers to the network structure of residual network,SEnet,and analyzes its characteristics.Combined with the characteristics of its own data set,it redesigned the number of convolution layers and the number of convolution kernels,trained a network model suitable for the separation of beverage commodities,and achieved the ideal classification effect,and calculated the parameters of the network model.In order to optimize and accelerate the analysis,a prototype system demonstration was finally carried out.In the study of some target detection algorithm and classification algorithm,in the light of the characteristics of data sets: drink goods bottles imaging fixed dimensions,detect the target transformation range is small,image annotation to spend time is long,take the advanced beverage goods target XingJian a two-step strategy and then classification and recognition,and to improve the network model to identify beverage goods.The final accuracy can meet the requirements of practical application,and the forward reasoning speed can also meet the requirements of real-time recognition.