A multi-class framework for fish species classification using deep learning technique.

Area of Science:

• Marine Biology and Ecology
• Artificial Intelligence in Fisheries
• Computer Vision for Biodiversity

Background:

• Accurate fish species recognition is vital for ecological studies, fishery management, and biodiversity preservation.
• Manual identification methods are inaccurate and costly, highlighting the need for automated solutions.
• Deep learning, particularly Convolutional Neural Networks (CNNs), shows promise but faces challenges with complex environmental factors and limited datasets.

Purpose of the Study:

• To develop and evaluate a deep learning system for automated fish species recognition using the YOLO object detection paradigm.
• To address the limitations of manual identification and existing automated systems in accurately categorizing fish species.
• To assess the performance of a proposed YOLO architecture on the Fish-Pak dataset, comparing it with other YOLO versions.

Main Methods:

• Utilized the You Only Look Once (YOLO) deep learning framework for object detection.
• Trained and tested the model on the Fish-Pak dataset, comprising 915 images of 6 tropical fish species.
• Conducted experimental comparisons between the proposed YOLO architecture and YOLO v3 and v4 to evaluate performance.

Main Results:

• Achieved a high overall accuracy of 99% for fish species identification.
• Obtained a mean Average Precision (mAP) of 99.65%, indicating superior detection performance.
• Demonstrated top performance results compared to existing literature on fish species recognition.

Conclusions:

• The developed deep learning system effectively recognizes fish species with high accuracy.
• The YOLO paradigm, applied to the Fish-Pak dataset, offers a robust and efficient solution for automated fish identification.
• This advancement has significant implications for ecological monitoring, fisheries management, and marine conservation efforts.