Utilizing Deep Learning for Morphological Analysis of Fish Species in Large Databases

Abstract

With the use of Deep Learning algorithms, it is conceivable to identify the object characteristics in an image or 
video frame as a collection of landmarks. To regulate the size of the fish and the location of its mouth and fins, 
eight landmarks are aligned after object identification and image segmentation are completed to isolate a fish in 
an image. This study included four common Mediterranean fish species: Diplodus palazzo, Sparus aurata (sea 
bream), Dicentrarchus labrax, and Merluccius merluccius (cod fish). Fish farms are used to raise the first three of 
these species. Ichthyologists are thus particularly interested in tracking these fishes' morphological characteristics 
in their natural habitat, and the suggested approach can help with this. Convolution neural networks and OpenCV 
were used in Python and MATLAB applications to build the suggested approach. Furthermore, we offer a 
thorough analysis of the leading deep learning methods for monitoring fish habitat, such as segmentation, 
classification, enumeration, and localization. Additionally, we evaluate several DL approaches in the underwater 
fish monitoring domains and investigate publically accessible datasets. We also review some of the difficulties 
and possibilities in the new area of deep learning for analyzing fish habitats. This publication is intended to guide 
marine researchers who wish to gain high-level knowledge of DL, follow our detailed tutorial to build it for their 
applications, and observe how it is developing to support their research. Additionally, it is appropriate for 
computer scientists who wish to examine cutting-edge DL-based techniques for monitoring fish habitat.