Design and implementation of image classification method based on deep metric learning

Image classification is a critical area of research in computer vision and deep learning, with applications in fields such as image segmentation and medical tumor detection. Despite significant advancements, challenges remain in improving classification accuracy and generalization. This paper introduces a novel deep metric learning-based image classification method to address these challenges. The model combines a CNN-based feature extraction module with a metric learning module that maps input vectors into a space where inter-class distances are maximized and intra-class distances are minimized. We propose a novel metric loss function that integrates contrast loss and triplet loss to improve accuracy and generalization. An adaptive sampling strategy dynamically selects contrast pairs based on sample difficulty, further boosting performance. Experiments with ResNet50 and GoogLeNet demonstrate the effectiveness of the proposed loss functions and sampling strategies. Additionally, a weight_decay training strategy mitigates overfitting and stabilizes training. Multiple evaluation metrics are used to assess model performance, showing that the proposed method achieves high accuracy and robust generalization.