The classification of multiple food items within a single image remains a significant challenge in computer vision due to occlusion, lighting variations, and diverse food appearances. This paper proposes a novel approach using the ResNet-18 model, optimized via Optuna-based hyperparameter tuning, for multi-class food image classification. A dataset comprising 6000 images across 20 Indian food categories was utilized. The combination of ResNet-18’s residual learning and Optuna’s efficient hyperparameter optimization resulted in a model achieving 96% accuracy, outperforming traditional approaches like MobileNetV2, EfficientNet B0, and InceptionV3. The results demonstrate that intelligent optimization strategies can substantially enhance classification performance, even with lightweight architectures.