Meta-Learning Enhancements in Waste Classification: Leveraging ResNet-50 and MAML for Efficient Few-Shot Learning

The escalating global waste crisis necessitates innovative solutions for efficient, sustainable, and adaptive waste management practices. Traditional deep learning-based waste detection systems require a large number of annotated samples for training, which is both labor-intensive and time-consuming. Meta-learning, a field of machine learning, offers a promising avenue to address these challenges by equipping models to learn from minimal data and rapidly adapt to new tasks. This study presents an approach for image classification using ResNet-50 in conjunction with Model-Agnostic Meta-Learning (MAML). The methodology leverages the ResNet-50 Convolutional Neural Network (CNN) for feature extraction and integrates MAML for image classification. The meta-learning paradigm offered by MAML facilitates the acquisition of a well-optimized initialization, which can be rapidly refined, thus enhancing the efficiency of few-shot learning tasks. By initializing the parameters through MAML training on the Mini-ImageNet dataset, the proposed model demonstrates improved accuracy with minimal training iterations on the TrashNet dataset, which serves as the benchmark for evaluating the efficacy of this approach. The results indicate a notable improvement in generalization accuracy, highlighting the potential of combining ResNet-50 and MAML for efficient and effective waste classification with limited annotated samples.