CNN-based species recognition and counting system for multispecies seagrass

Abstract

Effective conservation techniques aim at maintaining coastal biodiversity, carbon accumulation zones, and overall marine health, with increasing focus on important seagrass ecosystems. Conventional species identification and counting methods require extensive fieldwork, are labor-intensive, and are highly error-prone. This paper proposes an automated method based on deep learning techniques for species identification and counting from underwater photos in seagrass multihabitats. In this paper, Convolutional Neural Networks (CNNs) automatically identify and categorize several seagrass species through underwater imaging. The approach includes a vast annotated dataset of seagrass photos containing images of various species from diverse habitats. The constructed model addresses issues related to water and light conditions and complex multi-species environments by using advanced image processing, multi-stage feature extraction, and custom-designed CNNs. The results reveal that these species are identified and counted more accurately than with conventional methods, regardless of the reporting and operational timelines associated with those methods. Reliability and scalability are also enhanced in these modern techniques. The results further support more efficient monitoring across broader areas with reduced resource requirements, proving advantageous for environmental assessments through the CNN framework. In addition, the innovation is ideal for marine biologists and conservationists, as it provides opportunities to integrate automated systems for continuous surveillance and real-time ecosystem monitoring of seagrass habitats.