Applications of Cut, Paste, Learn Synthetic Image Generation, and Convolutional Neural Networks in Marine Animals Classification

Abstract

Marine animal classification using computer vision is an important task for various applications, including monitoring biodiversity, conservation efforts, and fishing management. However, obtaining and labeling real-world images and videos of these marine animals can be challenging due to the difficulty of accessing and observing these animals in their natural environments. This study explored the use of synthetic images as a training dataset for marine animal classification through Cut, Paste, Learn (CPL) as one approach to generating synthetic images. This study investigates the use of different convolutional neural network (CNN) models for classifying marine animals including Lobster, Squid, Crabs, and Sea Urchins. The Cut Paste Learn image generation method was utilized to augment the training data, which combines image manipulation techniques by pasting foreground images of the classes randomly to a background image to generate synthetic images. Further, the performance of the different CNN models on a dataset of real-world images of marine animals were evaluated and compared the results to determine the most effective model for this task. The findings in this study may be employed for the use of CNN models in marine biology and the application of the Cut Paste Learn image generation method in other domains.