Randomized trial utilizes a novel vision-based framework to quantify feeding intensity in aquaculture, indicating significant advancements in feeding technology.
Key Points
The aim is to accurately quantify fish feeding intensity in industrial recirculating aquaculture systems to optimize feeding strategies and reduce waste.
Developed a hybrid vision-based framework integrating CNN for feature extraction and ViT for context modeling.
Incorporated LSTM to analyze temporal dynamics of feeding activity.
Constructed a dataset of largemouth bass under industrial RAS conditions with data augmentation to enhance robustness.
Achieved over 98% accuracy in classifying four levels of feeding intensity.
Outperformed conventional CNN-based methods.
Enabled real-time quantitative evaluation of feeding activity for intelligent feeding systems.