Efficient Food Recognition with CBAM-Enhanced MobileNetV2: Achieving 83.5% Accuracy on Food-101

Food recognition on mobile and edge devices presents a significant challenge due to the need for high accuracy under limited computational resources. In this paper, we propose a lightweight yet powerful architecture based on MobileNetV2, enhanced with Convolutional Block Attention Modules (CBAM) and trained with an advanced data augmentation pipeline. Our approach systematically improves the baseline MobileNetV2 top-1 accuracy on the Food-101 dataset from 76.3% to 83.51%, an absolute gain of 7.21%. Crucially, our model achieves this performance with only 2.27 million parameters, outperforming the much larger ResNet-50 (25.6M parameters, 82.4% accuracy) while being 11.3× smaller. We demonstrate that strategic integration of attention mechanisms combined with mixed-precision training and modern regularization techniques enables stateof-the-art efficiency for food recognition tasks. The complete source code and pre-trained models are available at https://github.com/TayyabAbbas802/Enhanced-Food-Recognition-with-CBAM-Augmented-MobileNe