Research on The Efficiency of Mixed Commodities Classification based on Deep Convolutional Neural Networks

Abstract

With the wide application of deep learning in the field of image classification, the application of this technology in daily life has become more popular. In this context, self-service shopping based on deep learning makes people's lives more convenient. This paper extends the classification and identification of fruit and vegetable classes, which in turn investigates the classification of mixed commodities. This study aims to compare the performance of three widely used models, VGG, DenseNet, and ResNet, on a mixed commodities dataset. This research focuses on evaluating each model's performance on this dataset and determines which is the most suitable for this task. The study found that the accuracy levels of these three models varied. The training set's accuracy reached 0.91 for VGG16, 0.86 for Densenet201, and 0.74 for Resnet50. The validation set's accuracy reached 0.84 for VGG16, 0.87 for Densenet201, and 0.77 for Resnet50. In this study, DenseNet has performed relatively well and is more suitable for the mixed commodities dataset. The experiment has shown that to fulfill the high-efficiency requirements of self-checkout, the model shortcut can effectively save time and prevent the phenomenon of overfitting. Based on repeated experiments, it can be concluded that the data in the dataset used for self-service shopping model training should highlight its features and reduce the influence of background noise and other noises that may affect model training.