When the Task of Traditional Medicinal Herbs Classification Meets Supervised Prototype Contrastive Learning

This paper aims to explore and apply supervised prototype contrastive learning and instance contrastive learning methods to address the task of herbal medicine classification. Traditional medicine relies heavily on herbal medicine as a vital element in its practice. with a long history and rich experience in disease treatment and prevention. With the extensive application and research of herbal medicine, its classification has become a key research direction. However, traditional herbal medicine classification faces challenges of sample scarcity and class imbalance, which limit the performance and accuracy of classification algorithms. To overcome these problems, we propose a method that combines supervised prototype contrastive learning and instance contrastive learning. Prototype contrastive learning(PCL) is a method that improves classification performance by learning compact representations of classes, while Instance-wise contrastive learning(ICL) enhances the robustness and generalization ability of classifiers by contrasting and emphasizing the differences between entities of different classes. By combining these two methods, we aim to improve the accuracy and robustness of herbal medicine classification tasks. We conducted a set of ablation experiments on a herbal medicine dataset. The experimental results demonstrate significant performance improvements in the herbal medicine classification task for both contrastive learning methods.