Bangladeshi local fish detection using deep learning techniques

Abstract

This study uses deep learning techniques to provide a novel method for local fish understanding in Bangladesh. The following six native fish species are represented in the extensive dataset that was painstakingly gathered: 'Channa punctata (Taki),' 'Anabas (Koi macch),' 'Puntius (Puti),' 'Amblypharygodon (Mola macch),' 'Batasio tengana (Tengra),' and 'Ompok bimaculatus (Pabda).' The dataset was deliberately chosen to guarantee inclusion and diversity of different fish species that are frequently seen in Bangladeshi seas. We used cutting-edge deep learning methods, such as "InceptionV3," "Xception," "ResNet50," "VGG19," and a specially created Convolutional Neural Network (or "CNN"), to train and assess the fish detection model. These algorithms were selected based on their effectiveness in picture recognition applications and their capacity to extract complex features from various datasets. After extensive testing and training, our findings show that 'InceptionV3' outperforms all previous algorithms, obtaining a remarkable accuracy of 98.51%. The exceptional performance of 'InceptionV3' highlights its effectiveness in identifying the distinctive features of native fish species found in Bangladesh, highlighting its potential for useful applications in fish species identification. This work not only adds an important dataset to the field, but it also emphasizes how important it is to select the right deep learning method for the given local environment. The accomplishment of 'InceptionV3' in this particular situation provides opportunities for the use of precise and trustworthy fish detection systems, which are essential for managing fisheries, tracking biodiversity, and promoting ecological conservation in Bangladesh's aquatic environments.