AlexNet Vegetable Recognition Algorithm Based On Gibbs Sampling And Residual Structure

In self-service supermarkets and retail industries,rapid identification of vegetables is of great significance for promoting the commercialization of the vegetable industry and accelerating circulation speed.At present,the application of vegetable recognition algorithms still faces challenges such as low recognition accuracy and long response time.In addition,current vegetable recognition methods are only developed for a limited number of categories,which affects the application of automatic vegetable recognition.Therefore,how to improve the accuracy,speed,and multi category of vegetable recognition is a key issue for its widespread application and promotion.This article focuses on the algorithm and application research of vegetable recognition,focusing on the above issues.The main research content is as follows:(1)A Gibbs sampling based image segmentation algorithm is proposed to address the issue of low accuracy in vegetable recognition.Firstly,assume each vegetable image as a two-dimensional grid,construct a probability model for vegetable images,set classification categories,and obtain the initial labeled image.Secondly,the posterior distribution is obtained from the probability model through Gibbs sampling,that is,the maximum posterior probability probability distribution matrix of image pixels.Based on this matrix,the random number is obtained by rotating the X-axis and Y-axis,and the corresponding coordinates of the random number are calculated as the coordinate values of the target pixel.By continuously iterating to obtain the set number of sample points,that is,the sample set.Then,based on the sample set,randomly select points as the center points of the cut image,and cut a fixed size local image of vegetables.Finally,predict the category through a voting mechanism.The experiment shows that the Gibbs sampling based image cutting algorithm proposed in this article improves the recognition accuracy of Google Net,Res Net,and Alex Net network models by 7.41%,5.55%,and 12.08%,respectively.(2)A convolutional neural network model with fast convergence speed is proposed to address the problem of low computational efficiency in vegetable recognition.This model is based on the Alex Net convolutional neural network model.This paper designs a residual structure between layers to reduce the loss of image feature information;BN algorithm is added after the convolution layer,and PRelu activation function is used to replace the original Relu activation function,which speeds up the convergence speed of Alex Net network model and improves the recognition accuracy;Adding a global maximum pooling algorithm before the classifier reduces the number of parameters and improves computational efficiency.By combining with the Gibbs sampling image cutting algorithm mentioned above,this paper proposes a vegetable recognition algorithm Gi RAlex Net based on Gibbs and residual structure.Through systematic experiments on 55 types of vegetables,it has been shown that the Gi RAlex Net algorithm further improves the recognition accuracy of vegetables,with an accuracy rate of 97.78%.(3)Implemented the mobile application of the algorithm in this article.The Gi RAlex Net model was obtained through training in the vegetable image library system.The Gi RAlex Net model was ported using the Py Torch Mobile framework,and the UI interface was designed and debugged on the mobile end.The generated.APK file was run and installed on the intelligent electronic scale,achieving the mobile application of the intelligent electronic scale.The feasibility of the algorithm in this paper on the intelligent electronic scale was verified through system functional testing.