Abstract: A food recognition method was proposed based on the improved adversarial erasing technique that incrementally obtain discriminative regions. The method could identify each discriminative region using Otsu algorithm and morphological operations， thereby reducing noise interference. To validate the effectiveness of the proposed food recognition method， the comparative experiments were conducted on the Sushi-50， ETH Food-101 and Vireo-172 datasets with the methods presented in other literatures. The experimental results demonstrated that the proposed method can effectively mitigate interference from complex backgrounds in food images， thereby enhancing food recognition performance. Compared to ResNet-50， the method improved the Top-1 and Top-5 accuracy on the ETH Food-101 dataset by 2.6 and 0.8 percentage point， respectively.