Visual Anomaly Detection Dataset Collection for Shrimp Pond Monitoring: Foam, Surface Change and Feed Accumulation

This record consolidates three open-access datasets selected as the training and validation foundation for a visual anomaly detection system applied to shrimp farming ponds. The first, IWHR_AI_Lable_Floater_V1 (Qiao et al., 2025, Scientific Data, DOI: 10.1038/s41597-025-04594-9), comprises 3,000 images captured under real inland water conditions, encompassing 23,692 annotated objects across complex lighting and surface reflection scenarios; its relevance lies in the representation of visual alterations at the water surface. The second, Foam Analysis v14 (Sevil, 2024, Roboflow Universe, CC BY 4.0), contains 370 images oriented toward foam instance segmentation. The third, Shrimp and Feed Detection v3 (Sem 8 AI, 2025, Roboflow Universe, CC BY 4.0), includes 233 images with shrimp and feed detection annotations compatible with YOLOv8 and YOLOv11. Together, the three datasets support the comparative evaluation of YOLO architectures for early identification of anomalous operational conditions in aquaculture ponds.