Research On Commodity Recognition Technology Based On Deep Learning

With the improvement of people's living standard and the increase of commodity types,consumers' demands for goods are also higher and higher.Among them,the variety of goods and the description of goods make the identification of goods more and more important.In this paper,the commodity image recognition and text recognition in commodity image are studied based on the deep learning method.Compared with the traditional image recognition and text recognition algorithm,the image and text recognition algorithm based on convolution neural network is analyzed,and the research method of this paper is proposed.1)In order to reduce the influence of noise on the accuracy of image recognition,this paper applies the deep residual shrinkage network to the recognition of commodity image.The algorithm model is mainly based on the deep residual network,and then the soft threshold function and attention mechanism are integrated into the network model.Among them,the soft threshold function can set the unimportant feature noticed by attention mechanism to 0,so as to reduce the interference of noise information and improve the accuracy of commodity image recognition.The data set containing 51 kinds of goods was trained in the experiment,and the deep residual shrinkage network was compared with RESNET,alexnet and senet network respectively.The experimental results show that the depth residual shrinkage network can not only improve the accuracy of commodity image recognition,but also improve the speed of the model.2)This paper presents a text detection and recognition model based on ctpn + crnn.Firstly,ctpn text detection model is based on vgg16.With the help of text construction algorithm,it trains 10048 commodity image data sets.Firstly,the large text line is divided into small text boxes for feature extraction,and then the text box detection is realized by using side refine algorithm.Through the analysis of training and test results,the algorithm has a good effect for text box detection.Then crnn text recognition algorithm model is used to train and test 858750 Chinese and English text images.The algorithm uses two-way LSTM,combines the information about the text to achieve speculative recognition.At the same time,the algorithm combines the CTC loss function,which has its unique advantages for text recognition with variable length.Finally,the text detection and recognition are combined to realize the end-to-end text recognition of commodity image.Through the analysis of the experimental results,the accuracy of text recognition is about33%,the model can recognize the text content and realize the function of text recognition,but the effect needs to be improved,and the next work is expected to improve the accuracy of recognition.3)In order to describe the functional characteristics of the training system of commodity recognition model as a whole,and to sort out a set of software design ideas,implementation steps,view deployment and various technical solutions for program developers,this paper implements the training system of commodity recognition model.Through the idea of modularization,the system is convenient for system expansion and software testing.At the same time,users can call the system through the interface to realize their own commodity recognition model training.