Optimizing Deep Convolutional Neural Networks with Progressive Unfreezing for Enhanced Sports Activity Recognition

This study presents a methodology to enhance sports activity recognition using advanced deep Convolutional Neural Networks (CNNs). The objective is to fine-tune CNN architectures for accurate classification of various sports activities from image data. A systematic preprocessing pipeline employs data augmentation techniques-such as random rotations and horizontal flipping-alongside normalization and resizing for consistent training. This study explores several CNN architectures, including EfficientNetB0, ResNet50, EfficientNetB3, Xception, and EfficientNetB7, using a progressive unfreezing approach. The proposed appraoch achieved an impressive accuracy of 87.00% with EfficientNetB0 on the benchmark Clever dataset. This research contributes to deep learning methodologies in sports classification, showcasing the effectiveness of progressive unfreezing to improve accuracy and encouraging further exploration of diverse sports activities.