Detecting and classifying media images of athletes using convolutional neural networks – case study: Individual sports images

Abstract

Sports image classification using neural networks and machine vision is a rapidly expanding field, with applications in highlight reel creation, performance analysis, and illegal play detection. We present an innovative structure for sports image classification using convolutional neural networks (CNNs) based on deep learning in this article. Boxing, gymnastics, swimming, tennis, and weight lifting are five distinct sports that all fall under the umbrella of individual games. In terms of setting and attire, these various forms of athletic competition are very comparable. Specifically, the suggested deep learning model consists of 20-layers, among them, there are four CNN layers. The results show that the proposed model achieved a significant result in terms of accuracy, although the selected sports have similar characteristics to each other. For instance, boxing classification accuracy was 90.63%, gymnastics accuracy was 86.88%, swimming sport image classification achieved 94.06% accuracy, tennis classification accuracy is 88.13%, and the weight lifting was 89.06%, in the testing phase. The obtained results prove that the developed new sports image classification method is effective enough and has been improved.