Detecting LED Colors from Underwater Video Images for Visible Light Communication in a Fish Farm Monitoring System

This letter proposes an underwater visible light communication system using low‐cost devices for real‐time fish farm monitoring. The system transmits data using four‐color shift keying with blue, red, green, and white LED lights. To decode the signals from captured video, in this letter, a two‐stage method is proposed: YOLOv8 detects LED positions, and ResNet18 classifies their colors. Experimental evaluations conducted across four distinct underwater environments demonstrate high detection and color estimation accuracy under various turbidities and distances. Furthermore, the processing speed reaches 69 fps, confirming its capability for real‐time aquaculture applications.