FishFusionNet: An ensemble-based approach for fish species identification using SqueezeNet and efficient attention-boosted semi-local network

Accurate fish species classification is essential for biodiversity monitoring, fisheries management, and ecological research. Traditional identification techniques depend on manual examination, which is not only time-intensive and laborious but also susceptible to mistakes, particularly when distinguishing species with high intra-class similarity. To address these challenges, this study proposes FishFusionNet, an ensemble model combining SqueezeNet and the Efficient Attention-Boosted Semi-Local Network (EASLN) to achieve a balance between computational efficiency and robust feature extraction. SqueezeNet ensures a lightweight architecture with minimal computational overhead, while EASLN enhances fine-grained recognition through adaptive attention mechanisms. The proposed model is the first work to be evaluated on the BD-Freshwater-Fish dataset, achieving 97.74% accuracy, surpassing state-of-the-art deep learning models. Additionally, it was validated on the BDFreshFish dataset, where it attained 95.39% accuracy, demonstrating strong generalizability. The results highlight the effectiveness of FishFusionNet in automated fish classification, offering a scalable and efficient solution for ecological applications and sustainable fisheries management.