Research On Classification Method Of Jujube Appearance Quality Based On Channel Weighting And Information Aggregation

Jujube is a common medicinal food in China and an important agricultural export product.However,during the process of planting,picking and transportation,jujube are susceptible to damages such as insect pests,moisture and external mechanical forces.Therefore,classification according to the quality of jujube is of great significance for the subsequent storage and processing of jujube.At present,China's jujube quality classification method is mainly manual screening,which has the disadvantages of low efficiency,high cost and high labor intensity.In order to meet the needs of the agricultural market,finding an efficient and non-destructive automated classification method has become the current direction for the classification of agricultural products.In order to improve the accuracy of the jujube appearance quality classification,we propose a classification method of jujube appearance quality using convolutional neural network based on channel weighting and information aggregation.Firstly,the jujube sample is imaged by setting up a high-bright environment.The collected data is preprocessed and the region of interest is selected.The photos obtained after preprocessing are classified and divided into training sets,validation set and test sets.The design of the convolutional neural networks based on channel weighting and information aggregation,based on a Resnet-34 network structure,incorporates a squeeze and excitation module.It is designed to improve the representation of the network by modeling the dependencies of each channel,enabling network-selective learning to enhance the functionality of useful information and suppressing useless functionality by using global information.The integration of the information aggregation module allows the convolutional neural network to gather as much channel information as possible without increasing the complexity of the network,which increases the rich extraction of image features.At the same time,the batch normalization layer and the PRe LU activation layer are used in the network structure to accelerate the convergence rate of the network.Using the constructed data set to train the improved convolutional neural network,then perform a performance test on the trained model.The results show that the model can accurately predict four types of jujube: full jujube,dried jujube,cracked jujube and broken jujube,showing that the model performs well in the classification of the appearance quality of jujube.