Enhancement and classification of underwater species images using deep learning

Abstract

Automated classification of Marine species plays a significant role in studies dealing with a species count for population evaluation, behavior analysis, monitoring of the ecosystem, realizing the association between species and ecosystem, etc. Most marine species appear identical to human perception but differ in shape, structure, and color. Extracting those similar features from a degraded species image is challenging for traditional vision techniques. These minute-variation feature scan be extracted efficiently with the help of deep learning techniques. Hence, there is a need for automated techniques that can consistently classify marine species using deep learning in underwater videos without human interaction. This will lead to the production of knowledge on fish species and their habitats. The video and image information of marine species is acquired from underwater survey equipment such as Remotely Operated Vehicle (ROV), Autonomous Underwater Vehicle(AUV), etc. The underwater captured marine species images still suffer from image degradations, leading to the misclassification of species. Hence, the primary objective of this thesis focuses to develop effective image visibility improvement techniques for degraded underwater species images and to develop a robust classification-based network on computational intelligence approaches. To achieve this, the proposed research works were carried out in a series of modules namely: Data Acquisition module, Degraded Image Visibility Improvement module, and Deep learning species classifier to recognize the marine species images. newline