Published Paper Details:

Fish species identification via machine learning is critical for fisheries management, biodiversity assessment, and automated ecological monitoring. In this work, we employ a two stage pipeline: first, images are preprocessed by contrast normalization (CN) and Unsharp Mask Filter (UMF), and individual fish instances are detected using a YOLOv5 based model. In the second stage, detected instances are classified into species using a fine tuned Convolutional Neural Network (CNN), specifically ResNet50, where NsN_sNs denotes the number of target species. Data augmentation--including rotation, scaling, and color jitter--is applied to improve robustness to underwater lighting variability. Models are trained and evaluated on a dataset of MMM labeled fish images (where MMM defines the total image count), achieving an overall classification accuracy of 98.7% and mean F1F_1F1-score of 0.985. No reference citations appear in this abstract. All variables--CNCNCN, UMFUMFUMF, NsN_sNs, MMM, and F1F_1F1--are defined.