Variational quantum enhanced deep transfer learning for small underwater aqua species image classification.

Precise underwater classification of small aquaculture species is essential for sustainable fisheries management, biodiversity monitoring, and automated marine ecosystem analysis. But it is still a challenging task owing to underwater image distortions from poor visibility, lighting changes, occlusions, and the high computational complexity of traditional deep learning models. To address these issues, we propose a Lightweight Variational Quantum Enhanced Deep Transfer Learning framework. This hybrid deep transfer learning model integrates pretrained classical convolutional neural networks with variational quantum circuits to improve feature representation and classification efficiency. The framework is designed to reduce computational complexity while enhancing accuracy by leveraging quantum feature extraction techniques. Experimental evaluations on curated small aquafarming species dataset demonstrate that the proposed approach achieves high classification accuracy (up to 99.25%) with significantly fewer parameters and floating-point operations, indicating its potential for resource-constrained applications. Ablation studies further validate the impact of quantum layers on model performance. These results suggest that quantum deep transfer learning models can offer a promising direction for robust and efficient underwater species classification.