Recognition and localization of human hand acupoints based on HrNet

With the rapid development of image recognition and optical imaging technologies, there has been significant progress in the field of acupoint localization. Therefore, in this paper, a deep learning-based human hand acupoint recognition and localization method is proposed. Aiming at the problems of large positioning error and weak algorithm generalization ability of the existing intelligent acupoint localization methods, we carried out the following research: firstly, we independently constructed a human hand acupoint dataset containing 948 high-quality hand images and labeled thirty hand acupoint points based on the national standard. Secondly, to address the problem of poor accuracy of hand acupoint point taking, this paper proposes an acupoint localization model based on the hrnet model. The model replaces the commonly used 2D heat map detection head with 1D Simcc detection head to effectively solve the problem of quantization error. And the Focus module, CBAM attention mechanism, and TVConv module are introduced in the feature extraction part. Compared with the original HrNet model, the AP value of the improved acupoint localization model rises by 2.6% and the PCK value rises by 1.1%. The improved model enhances the feature extraction ability of the network effectively improves the accuracy of acupoint point detection.