THE BIOLOGICAL IMAGE TECHNOLOGY FOR METABOLIC ENZYMES IN ATHLETES’ ACCELERATION ABILITY OF COMPETITION BY DEEP LEARNING

With advances in biotechnology, investigating the effects of various substances on the human body has become a foundation for developing scientific training methods to enhance athletic performance. Progress in artificial intelligence (AI) and deep learning (DL) has enabled more precise acquisition and analysis of biological data. This study addressed the challenge of indistinguishability and overlap among cells in biological images by focusing on accurate segmentation of cellular images to analyze the distribution of metabolic enzymes. Specifically, this study targeted the problem of precise quantification of metabolic enzymes in sports biology and optimized a DL framework based on a ResNet (Residual Network) encoder and a U-Net decoder (Res-UNet). To address the inherent challenges of metabolic enzyme imaging, tailored structural modifications, including dilated convolution strategies, were introduced, and the segmentation performance of the network was systematically validated and evaluated within this specific application context. Experimental results demonstrated that the proposed segmentation method improved average accuracy by 0.218 compared with the original U-Net. Through iterative training and data preprocessing strategies, the improved network achieved superior recognition accuracy on the test set. When combined with athlete performance data, the findings confirmed a positive correlation between metabolic enzyme content and sprinters’ acceleration ability. Accordingly, sprint training should be designed with consideration of athletes’ distinct metabolic systems and enzyme production rates. Furthermore, the arrangement of speed and endurance training should integrate the development of maximum velocity, acceleration capacity, and strategies tailored to different sprint force modes. Overall, this study provides a theoretical basis and practical reference for sprint training, with potential contributions to breakthroughs in sprint performance.