With the rising global demand for aquatic products, aquaculture has become a cornerstone of food security and sustainability. This review comprehensively analyzes the application of deep learning in sustainable aquaculture, covering key areas such as fish detection and counting, growth prediction and health monitoring, intelligent feeding systems, water quality forecasting, and behavioral and stress analysis. The study discusses the suitability of deep learning architectures, including CNNs, RNNs, GANs, Transformers, and MobileNet, under complex aquatic environments characterized by poor image quality and severe occlusion. It highlights ongoing challenges related to data scarcity, real-time performance, model generalization, and cross-domain adaptability. Looking forward, the paper outlines future research directions including multimodal data fusion, edge computing, lightweight model design, synthetic data generation, and digital twin-based virtual farming platforms. Deep learning is poised to drive aquaculture toward greater intelligence, efficiency, and sustainability.

This study presents a lightweight RGB vision-based express-parcel detection framework for complex logistics environments. An Express Parcel Recognition (EPR) dataset is first constructed from RGB images acquired through multi-source data collection, followed by unified re-annotation and Active Learning-based pseudo-labeling to reduce manual annotation effort while maintaining label consistency. Based on YOLOv11s, a lightweight detector termed EPR-YOLO is developed by integrating SFAConv and NGAM to strengthen multi-scale feature extraction and global semantic modeling from RGB visual inputs while controlling computational complexity. Quantization-Aware Training is further introduced to obtain an INT8-quantized model with reduced storage requirements and limited accuracy degradation, providing a basis for subsequent lightweight deployment. Comparative experiments, ablation studies, and independent test set evaluations demonstrate that EPR-YOLO achieves an mAP₅₀ of 71.654%, improving the YOLOv11s baseline by 0.854 percentage points while reducing the parameter count by 9.5%; the quantized model further retains most of the detection capability of its floating-point counterpart. In addition, a FastAPI-based Web application is implemented to support RGB image acquisition-based single-image and batch detection, result visualization, and online inference. The proposed framework therefore provides an integrated sensing-to-deployment solution covering RGB image-based parcel perception, dataset construction, lightweight model optimization, quantized deployment, and application-level validation for intelligent logistics systems.