Under Water Objects Detection and Classification using Deep Learning Technique

Abstract

Seas and oceans, vital sources of food and critical reservoirs of sustenance and ecological balance, present formidable challenges to exploration due to their profound depths. Addressing this, the integration of deep learning with autonomous robots emerges as a pivotal solution for the real-time detection and classification of underwater objects through the analysis of images captured in challenging conditions. Our key objective of this research is to find an impressive deep learning method that can be useful for underwater objects detection and classification in challenging conditions. This paper talks about making deep learning models better for underwater places. It deals with problems like low light, blurry images, and noisy environment in the water. Focusing on “The Brackish Dataset” our research leverages the YOLOv8 models, augmented with the MaxRGB filter, to robustly detect and classify underwater entities. We use MaxRGB filter in our experiment to increase the content separability in images. A thorough performance evaluation, including precision, recall and comparison with state-of-the-art models from prior work, proves the effectiveness of our approach. Our detector archives 98.6% mean average precision and classifier gains 99.6% classification accuracy. The findings not only affirm the suitability of YOLOv8 for underwater exploration but also highlight its potential strength in diverse fields, such as marine resource identification, rescue operations and ecosystem preservation. The intersection of deep learning and underwater environments opens new avenues for technological advancements with far-reaching implications for both scientific research and practical applications.