Shot Analysis of Batsmen in Cricket Matches using Transfer Learning Techniques

It is now possible to use deep learning techniques in a variety of disciplines thanks to advancements in hardware technology. Deep learning's Convolutional Neural Network (CNN) design has completely changed computer vision. One area where the use of machine vision is flourishing is in sports. Cricket is a challenging sport that involves many various kinds of shots, bowling motions, and other actions. In a cricket match, each batsman uses a unique batting motion. We use this idea to categorize the various kinds of strokes a batsman takes. In this paper, we suggest a CNN model that uses transfer learning to identify seven distinct cricket batting shots based on batting actions. To train the suggested model and assess its accuracy, we used a dataset called Shot-Net dataset, which contains 4900 images of these seven batting strikes. To construct our model, we have used an array of transfer learning techniques that were trained previously on Image-Net dataset. We chose the algorithm that had the greatest efficiency and learning potential. After experimenting with various approaches, we discovered that the greatest level of model accuracy was achieved by eliminating the weights of 17 layers of the VGG-19 network, expanding the architecture, and training the entire developed model. On the test dataset, our approach achieved an accuracy of 98.61% and minimal cross-entropy loss.