Deep Learning-Based Action Recognition and Technique Classification in Cricket Using CNN Models: A Study on the Cricket Shot Image Dataset

Abstract This study explores a deep learning-based framework for classifying cricket batting shots using Convolutional Neural Networks (CNNs). The goal is to identify and categorize common shot types such as drive, pull, sweep, cut, and lofted shots from still images. Using the Cricket Shot Image Dataset available on Kaggle, a custom CNN architecture was trained and evaluated for image-based action recognition. The proposed system employs convolutional feature extraction, ReLU activation, max pooling, and fully connected layers to learn discriminative motion patterns from image frames. The model achieved an overall classification accuracy of 96.2%, outperforming baseline approaches such as VGG16 fine-tuning and Mobile Net transfer learning. Experimental results indicate that CNN-based representations can effectively capture spatial and pose-related cues relevant to cricket shot classification. This work demonstrates the potential of computer vision in technique evaluation and player performance analytics for sports coaching and digital broadcast applications. Keywords: Cricket analytics, Action recognition, Convolutional Neural Network, Image classification, Sports AITop of Form Citation E. Jesudin Rajesh “Deep Learning-Based Action Recognition and Technique Classification in Cricket Using CNN Models: A Study on the Cricket Shot Image Dataset” International Journal of Current Science Research (IJCSR) e-ISSN: 2454-5422: 12(4): 2026: 1 - 5