Fishnet28:

Abstract

As urbanization advances, technological integration has become a necessity for the sustainable management of diverse ecosystems, where the fish world or the aquatic realm is no exception. This study introduces "FishNet28," an approach aiming to classify Bangladeshi indigenous fish species and decrease the information gap in Bangladesh. The primary objectives of FishNet28 encompass dataset construction, accurate species classification, custom machine learning model development, and the creation of a user-friendly mobile application, addressing critical gaps in fish identification methodologies and knowledge dissemination systems. At present, many indigenous fish species in Bangladesh are on the verge of extinction due to overfishing or false identification. Our methodology for FishNet28 addresses these fish identification gaps by involving the collection of a large dataset of fish images and classifying them with the help of machine learning and deep learning techniques. This study compares the performance of four pre-trained models: DenseNet201, Xception, ResNet50V2, and InceptionV3. Each model was classified, and their performance metrics were recorded for comparative analysis. Additionally, a custom layered DenseNet201 model was developed and evaluated. The custom FishNet28 model demonstrated superior performance with a training accuracy of 99.53% and a validation accuracy of 99.86%, highlighting its potential for accurate species classification in practical applications.