Dry Fish Image dataset: Data-driven analysis and deep learning-based classification

Area of Science:

• Computer Vision
• Machine Learning
• Deep Learning
• Fisheries Informatics

Background:

• Data-driven research in computer vision requires diverse, real-world datasets.
• Existing datasets may not capture the complexities of food items like dry fish in various market conditions.
• The need for specialized datasets to advance applications in food recognition and supply chain management is growing.

Purpose of the Study:

• To present a comprehensive dry fish image dataset framework for computer vision and machine learning research.
• To support data-driven investigations into dry fish classification, feature extraction, and related applications.
• To facilitate advancements in automated food recognition, market automation, and supply chain digitization.

Main Methods:

• Collected high-quality RGB images of twelve dry fish species from multiple markets in Dhaka.
• Acquired images using mobile cameras under natural lighting, incorporating variations in background, angles, and obstructions.
• Performed expert-verified manual classification, duplicate removal, and format standardization for data integrity.

Main Results:

• A diverse dataset of 12 dry fish types, reflecting real-world variations in appearance, handling, and presentation.
• Images captured under natural lighting with variations to enhance data diversity for practical deployment scenarios.
• A structured dataset, preprocessed for immediate use with deep learning frameworks and further expansion.

Conclusions:

• The dry fish dataset provides a valuable resource for computer vision, machine learning, and deep learning applications.
• It enables research in areas such as image classification, feature extraction, data imbalance analysis, and explainable AI.
• The dataset supports advancements in low-resource food recognition, market automation, supply chain digitization, and fisheries informatics.