A novel approach for satellite coastal-aquaculture cage and raft detection using sensor band misregistration parallax over coastal waves

Keywords:  coastal aquaculture, cage and raft aquaculture, parallax effect, deep learning, Sentinel-2, PlanetScope
Abstract Type: Paper Abstract

Abstract

Coastal aquacultural production has increased significantly in the last several decades with important economic and environmental impacts. Cage and raft aquaculture (CRA) is an established near-shore method of cultivating aquatic species by enclosing fish or shellfish in floating or supported metal cages, or suspending lobsters and seaweed from wooden rafts. CRA satellite monitoring is complex because the cages and rafts may be fully or partially submerged, they may be made with a variety of materials, have different spatial arrangement, and due to variable ocean surface conditions. In this study, a novel method to map CRA by leveraging parallax effects present in multi-band satellite imagery is described. Parallax arises from detector orientation and acquisition time difference among spectral bands that cause moving objects to appear at slightly different locations among bands. We show that the parallax effect can be captured by precise dense matching of water surface features among spectral bands. The resulting parallax maps are used to map wave and ocean current movements that refract around stationary aquaculture structures, enabling their detection against the dynamic background of surrounding water. The methodology is demonstrated in coastal regions of China using a U-Net deep learning model with multi-temporal Sentinel-2 (10m) and PlanetScope (3m) reflectance, and a temporal composite of the parallax magnitude. The method improves CRA mapping over using just reflectance and has potential for large-scale mapping of coastal aquaculture facilities.