Comparative Study of Lightweight Deep Learning Models for Soccer Penalty Kick Image Classification

Image classification plays a crucial role in sports video analysis, providing insights into player performance and tactics. This study classifies soccer penalty kick images into target zones using lightweight deep learning models. A dataset of 3,000 images from public match videos was divided into three penalty kick regions. Four convolutional neural networks-ShuffleNet-v2, MobileNet-v3, EfficientNet-b0, RegNetX-400MF-and a Simple Vision Transformer were evaluated. Results show all models performed competitively, with EfficientNet-b0 achieving the highest accuracy of 84.11%. These results highlight the effectiveness of lightweight networks for soccer image classification and their potential in real-time sports analytics.