A Novel Deep Learning Framework for Food Ingredient Classification Using ResNet and SMO Classifiers

The research proposes a new deep learning (DL) system that uses ResNet to extract features from food images and Sequential Minimal Optimization (SMO) classifiers for prediction. In addition to having trouble handling complex, high-dimensional datasets, traditional machine learning (ML) methods like Support Vector Machines (SVMs) also have trouble operating in real time. The proposed approach combines ResNet's capacity to learn hierarchical features with SMO's robust classification capabilities, delivering both accuracy and scalability advantages. The proposed method performs better than existing method, according to the testing results, with a maximum accuracy of 93.4%, precision and recall of 91.7% and 92.5%, respectively, whereas the existing system achieves only 85.3% and 78.5%. While it has longer trained times (7.5 hours), it has faster inference speeds (88 milliseconds). Robustness against noisy data (88.7%) and low-quality data (89.5%) contributes to the proposed system's strength, allowing for performance optimization that is excellent for real-time applications such as food safety and nutritional monitoring.