Automatic Quantification of King Salmon Farm Net Occlusion Via Biofouling Detection

Biofouling, the growth of organisms on submerged structures, poses significant challenges and costs for the aqua-culture industry. Implementing an image-driven approach to net cleaning, based on the actual biofouling growth rather than predetermined schedules or equipment availability, can offer substantial benefits. However, existing image analysis methods require manual processing, making the procedure highly time-consuming. This paper proposes an automatic process for analyzing biofouling images and quantifying net occlusion without human intervention. The study focuses on two types of bio-fouling images from king salmon farms in New Zealand, i.e., experimental net panels with blue backgrounds and GoPro video recordings of larger net sections. A novel image analysis method, incorporating adaptive thresholding and color discrimination, is proposed for processing images with blue backgrounds. For the GoPro images, we develop an automated analysis method to handle varying image quality. The proposed approaches can accurately detect net occlusion across two kinds of biofouling images, significantly improving the speed and effectiveness of biofouling assessments.