Seaweed aquaculture has myriad benefits running the gamut from meeting increasing seafood demands to enriching ecosystems by producing more oxygen-rich waters around them. Taking into account the current surge of interest in eco-friendly fuel sources, kelp has the potential to replace gasoline, if cultivation can be efficiently scaled up. Kelp farms have been shown to boost marine biodiversity in nearby waters similar to kelp forests, so kelp cultivation could benefit local wild populations and potentially fisheries as well, through added habitat, but the tools to understand these relationships in situ remain underdeveloped.
Differentiating marine species from kelp can be efficiently achieved by applying object-detection techniques to underwater visual data. Neural networks have yielded promising results in numerous fields and applications, including natural language processing and medical imaging. Object detection in videos is one of the more popular applications, for which the current industry standard is the You Only Look Once (YOLO) algorithm. YOLO is a convolutional neural network which detects objects in real time and assigns probabilities of likelihood to the detected objects. Prior literature has used YOLO-based neural networks for object detection, although not for assessing the impact of seaweed aquaculture on marine biodiversity.
Our aim was to classify fish and kelp in underwater video data using the YOLOv5 model. Video data obtained from a test farm in the Santa Barbara channel were captured by the UCSB Coastal Oceanography and Autonomous Systems Lab. Because the fish do not stand out in the video frames as clearly as the kelp, we made adjustments to a large pre-trained YOLOv5 model and retrained the model using sampled video frames from the kelp farm. Figure 1 shows example results from testing the trained model. Through this work the network’s confidence levels have been improved and the technique has been applied to multiple species of fish to identify which species are most commonly found associated with seaweed farms. Classifying fish and kelp provides a better understanding of the biodiversity near kelp farms, increasing the importance of kelp farms to the marine ecosystem, and should in turn inform management of sustainable farming practices.