A Real-Time Fish Recognition Using Deep Learning Algorithms for Low-Quality Images on an Underwater Drone
摘要
The maritime industry employs various tools, including underwater drones, to search for and monitor marine life. The primary objective of this study was to contribute to the preservation of the natural environment by investigating the species of fish present in different bodies of water. The research used three deep learning algorithms: Haar Cascade, Single-Shot Multibox Detector, and inception method for automatic fish detection. The study targeted Clark’s Anemone fish, Moorish Idols, and Blue Devils for species detection. Through simulations, the three approaches demonstrated high precision and efficient recognition times, achieving a maximum accuracy of 73% for fish recognition and 58% for species detection.