Academic Thesis

Hilsa fish (Ilish) is a crucial part of the Bangladeshi diet and economy, but accurately  distinguishing Healthy Hilsa from Jatka and visually similar species like Chandana and Gurta  presents a significant challenge in fish markets. To address this issue, we propose an  automated multiclass classification system using deep learning, with a novel hybrid model that combines VGG16 for feature extraction and a Graph Convolutional Network (GCN) for classification. This research uses a unique, self-collected dataset of Hilsa and similar species. The system is designed to differentiate Healthy Hilsa from other species (Jatka, Chandana,  Gurta, and Others), particularly overcoming the challenge of distinguishing Healthy Hilsa  from Jatka, which bears close visual similarities. This automated system  offers practical applications for fish market quality control and consumer authentication of Hilsa fish, helping to prevent Jatka overfishing and ensuring accurate identification of this  economically vital species.