Condiment recognition using convolutional neural networks with attention mechanism

Food adulteration is a signification food safety problem. Accurate identification of different foods is very important for the development of related food processing industries and food detection technology. In this study, CondimentNet was used to identify five kinds of food materials with similar appearance but different efficacy, such as fennel, cumin, caraway, Murraya paniculata and rosemary. Based on the original ResNet18 model, CondimentNet is mainly improved as follows:(1) An appropriate number of scSE attention modules are introduced. (2) Modified the size of the convolution kernel in the last residual module. (3) Modified the classifier structure. After pre-processing, the collected data is imported into CondimentNet for training and recognition. The experimental results show that the improved network recognition accuracy is 95.71 %, which is 1.11 % higher than the original resnet18 network. The above operation improves the recognition accuracy of the network without significantly increasing the training cost. In addition, compared with other advanced models, the superiority of CondimentNet network is verified. The classification of different varieties of spices by convolutional neural network verifies the feasibility of deep learning algorithms in the field of food detection, and promotes the development of identification technology of similar food raw materials. It provides a potential method for intelligent and accurate classification in the field of food.