An optimized deep learning approach for detection and classification of player actions in football game

Abstract

Player position detection and movement prediction is a hot research topic in sports. In this research work, the detection of players and classification of actions are performed using deep learning algorithms You Only Look Once (YOLO-V5) and Mayfly optimized deep neural network. The major contribution of this work is YOLO-V5, MixNet Convolutional Neural Network (MixNetCNN), and the mayfly optimization algorithm is applied for new domains like player actions in football games to show these techniques’ innovative performance. Here, the YOLOv5 (version 6.1) is used to accurately detect and classify the actions in football games. MixNet CNN is utilized to resolve the overfitting issue while classifying the actions of the football players. The mayfly optimization algorithm is utilized for optimizing the Boltzmann machine weights and parameters of classifiers helps to resolve the local optima issues. Experimentations on our dataset provide better prediction performance for recall, precision, f1-score, and mean average precision (mAP) metrics. The precision obtained by the proposed model is the maximum for all classes when the confidence score is 0.911. Comparative analysis with the existing approaches validates the better performance of the proposed detection and classification model.