# A Lightweight YOLO-PEGA-Based Method for Quantifying Fish Feeding Intensity

## TL;DR

This paper introduces a lightweight AI model to detect fish feeding intensity by analyzing water splashes, helping reduce feed waste and improve aquaculture sustainability.

## Contribution

A novel YOLO-PEGA-based model with improved small-splash detection and reduced model size for real-world aquaculture deployment.

## Key findings

- The model achieved 0.86 precision and 0.80 recall for splash detection on the validation set.
- The model size was reduced by 72.3% compared to the baseline while maintaining performance.
- The method supports real-time feeding decisions to reduce waste and improve water quality.

## Abstract

Feeding fish based on experience or a fixed schedule often fails to reflect how hungry the fish are at a given moment. Overfeeding not only wastes feed but can also worsen water quality. During feeding, fish compete for pellets and frequently disturb the surface, producing splashes of different strengths. These visible splashes provide a practical clue to feeding intensity. In this study, we built a labeled video–image dataset of splash events and trained an automated model to detect small and weak splashes more reliably. The results show that the method achieves high detection accuracy on validation data while using a smaller model that is easier to deploy on farms. By converting splash observations into quantitative information, the approach offers actionable support for deciding when to stop feeding, how long to feed, and how much feed to deliver. This can help reduce feed waste and lower the risk of water-quality deterioration, contributing to more efficient and sustainable aquaculture.

In aquaculture production, manual or fixed-schedule feeding often fails to match the real-time feeding level of fish schools, and overfeeding can lead to feed wastage and water-quality deterioration, which has become a major bottleneck for both large-scale farming efficiency and environmental sustainability. During feeding, intense competition and jumping behaviors generate splashes of varying magnitudes, which can serve as an indirect visual proxy for hunger intensity. In this study, we constructed a frame-level splash-annotated dataset and performed data preprocessing. Building upon YOLO11 pretrained weights, we introduced a P2–P5 four-scale detection head to enhance small-splash recognition, injected EGMA into the backbone C3k2 blocks, and replaced stride-2 downsampling convolutions with a three-branch ADown operator. On the validation set, the proposed YOLO11-PEGA achieved a precision of 0.86 and a recall of 0.80, with mAP@0.5 exceeding 0.80 and mAP@0.5–0.95 exceeding 0.30. Compared with the baseline model, the parameter count was reduced by 72.3%. The results demonstrate that the proposed model maintains stable detection and evaluation performance under complex environmental conditions, providing actionable decision support for feeding-threshold setting, feeding-time determination, and feed-amount adjustment.