An Improved Res2Net-Based Model for Classifying the Appearance of Deer Antler Slices

Deer antler slices are highly valued in Chinese herbal medicine due to their medicinal properties. However, the current process for classifying these slices is time-consuming and subjective. To overcome this issue, we propose an intelligent classification and recognition model based on the Res2Net architecture. Our neural network utilizes an inverse bottleneck structure to enhance grouped convolution and reduce model parameters and computation time. Additionally, we integrate an improved grouped convolution into the Res2Net model and leverage the efficient channel attention (ECA) mechanism to improve feature extraction. Our model achieves an impressive 97.96% accuracy in classifying deer antler slices and outperforms other related models. This approach can accurately differentiate between different types of deer antler slices and is particularly suitable for small-scale datasets.