Research On Vegetable Recognition Based On Deep Learning

With the rise of food safety traceability,unmanned supermarkets,and independent shopping,the automatic identification technology of vegetables and other agricultural products in circulation and sales has become an urgent problem to be solved.In the research process of image recognition technology,the research of vegetable image recognition technology has mainly experienced two stages based on traditional image processing and deep learning technology.The former relies on artificially defining vegetable features,such as color features,texture features,shape features,which has the disadvantages of complex recognition algorithm,high platform requirements,and difficult function expansion.The latter mostly uses convolutional neural networks and their improved algorithms,which can extract high-dimensional abstract image features and improve the accuracy and robustness of vegetable recognition.Therefore,based on the deep learning convolutional neural network model,this paper conducts relevant research on vegetable recognition technology and performs the following work.(1)Construct a vegetable dataset based on color and texture features.First,the national vegetable classification standard is used as the basic classification basis.Under the research background of this paper,a vegetable data set based on color and texture characteristics is constructed.Then,according to the basic steps of image preprocessing in the process of image creation,the image is optimized for visual quality and the number of image samples is expanded.Finally,in order to facilitate the reading and loading in the image recognition process,the label of the image is designed,mainly considering the packaging,the freshness of vegetables and other factors.(2)Research on vegetable recognition based on convolutional neural network.First,introduce the working principle and parameter training method of each layer of convolutional neural network.Then,based on the study of the difference in data input of the convolutional neural network,a shallow convolution parameter migration is initialized for training,and the deep convolution parameter is reinitialized for training.Finally,the simulation and experimental statistics of vegetable recognition were carried out on the Alex Net,VGG-16 model and its migration model.The vegetables with wrong recognition results were counted and the confusion matrix was drawn.(3)Research on confusion vegetable identification based on bilinear convolutional neural network.This chapter first introduces the model architecture and learning mechanism of the bilinear convolutional neural network,Then,based on the input of the bilinear convolutional neural network,a cluster constraint algorithm based on spectral clustering is proposed,which constrains the upper and lower limits of the clusters of vegetable categories in the confusion matrix,and uses the contour curve to determine the optimal cluster Value,the vegetable samples in the loaded cluster are input to the bilinear convolutional neural network model for further training and recognition.Finally,an experimental simulation of the proposed method is carried out.The experimental results show that the method can effectively improve the identification of vegetables that are easily confused.(4)Platform design based on X6818 vegetable identification.This chapter mainly designs the vegetable recognition system based on X6818.Firstly,it introduces the hardware structure and peripheral modules of X6818,and combines the raspberry pie wide-angle camera to build the hardware image acquisition and recognition hardware architecture.Secondly,build an embedded operating environment,a deep learning software system and its software application environment,implement vegetable image acquisition and recognition,design a QT interface to test the vegetable recognition system,the test results show that the platform can identify vegetables online.