Food Classification Model Based on Improved MobileNetV3

Food provides people with energy and nutrition. A scientific and reasonable diet can greatly guarantee a healthy life. Food classification technology is the fundamental work of research like diet health detection. Food classification has gradually become a research hotspot in artificial intelligence field. Currently, there exists several problems in the field of food classification such as lack of public datasets, large consumption of computing resources and low classification accuracy, which are difficult to deploy on portable devices. Based on the actual needs, this paper aims at the problems in the field of food classification mentioned above and has completed the following works: Building a 102-type food dataset containing 72815 pieces of samples; Experiments were carried out based on this dataset and mainstream neural networks. Introduce Coordinate attention mechanism to improve MobileNetV3 neural network and improve classification accuracy.