Fine-Grained Image Classification Model M-ResNet50 Based On Small Collar Dataset

Within this manuscript, a refined model, MResNet50, which builds upon the ECA-ResNet50 framework, is suggested for the nuanced classification of collar images, complemented by the introduction of the MC-Loss function. Because the collars' inter-class differences are concentrated in a small area of the image, traditional classification networks are difficult to distinguish effectively. The M-ResNet50 model significantly improves classification performance by focusing on discriminative and discriminative regions. Experimental results on the Collar-6 private dataset show that M-ResNet50 not only outperforms several classical and novel classification models, but also has better separability of extracted features. In addition, MResNet50's experiments on the public dataset DeepFashion also verified its effectiveness in the field of clothing image classification, indicating that the model is not only suitable for collar image classification, but also can be extended to a wider range of clothing image classification tasks.