Enhancing Aquaculture Net Pen Inspection: A Benchmark Study on Detection and Semantic Segmentation

Abstract

The aquaculture industry is critical in global food production, with net pens being a vital component in fish farming operations. Regular inspection of these net pens is essential to ensure their structural integrity, prevent fish escapes, and monitor biofouling. However, manual inspections are time-consuming, labor-intensive, and subject to human error, driving the need for automated solutions. Detection and segmentation are computer vision techniques that offer a promising approach to automating these inspections by enabling precise identification and classification of various components within underwater images. This paper presents a novel dataset designed specifically for detection and semantic segmentation in the context of aquaculture net-pen inspections. The dataset comprises diverse high-resolution underwater images and annotated with multiple classes, including net holes, biofouling, and vegetation. We also provide a benchmark evaluation of state-of-the-art detection and semantic segmentation models using standard performance metrics. We evaluate their benefits both qualitatively and quantitatively in aquaculture inspection. As a result, we recommend using the YOLOV8 model for the detection and segmentation task, as it offers an optimal balance between performance and computational efficiency, making it well-suited for real-time inspection. The dataset and the detection pipeline provide promising opportunities for further research in aquaculture net-pen inspection tasks.