Design Of Fruit And Vegetable Electronic Scale System Based On Deep Learning

At present,supermarket and agricultural trade market are developing towards the direction of intelligence,unmanned and convenience,which can not only reduce the sales cost of supermarkets and agricultural trade markets,but also improve the sales efficiency.However,fruit and vegetable sales hinder the development of supermarkets and agricultural markets.In the fruit and vegetable sales zone,staff still need to identify commodity categories,input commodity numbers,weigh and print labels.Therefore,it is of great significance to study the classification and identification methods of fruits and vegetables and design an electronic scale which can automatically identify the fruits and vegetables.Based on the lightweight neural network MobileNetV3-Small,this paper proposes a new fruit and vegetable classification algorithm FV-MobileNetV3 S,and compresses the model by weight quantification.Finally,a fruit and vegetable classification and electronic weighing system is designed based on raspberry pi 4B.The model of general deep learning algorithm is very large,it is difficult to run in embedded system or slow to occupy large resources.Based on the lightweight neural network MobileNetV3-Small,aiming at improving the accuracy and running speed,this paper adopts Dense Net connection algorithm to optimize the backbone extraction network structure.The SENet attention mechanism is improved,and a new attention mechanism GM-ECANet is proposed,which makes the algorithm more robust;finally,an improved MobileNetV3-Small classification of fruits and vegetables is established The algorithm,FV-MobileNetV3 S,is verified by experiments on fruits and vegetables data set and standard data set CIFAR-10,which proves the feasibility of the algorithm.In order to make FV-MobileNetV3 S algorithm run reliably on embedded platform,this paper studies the model compression technology.Through comparative analysis,the weight quantification method is selected to expand the model compression,and implemented on raspberry pi 4B.The experiment shows that FV-MobileNetV3 S algorithm has good recognition effect after model compression.Based on the hardware of raspberry pi 4B development board,Py Qt5 and My SQL software,a set of electronic weighing system for fruit and vegetable classification was designed,including database management and human-computer interface.The running test results shows that the development system realizes the expected function,and also verifies the correctness and feasibility of the research method.