Underwater <i>Phaeocystis globosa</i> monitoring based on ACE–DCP enhancement and YOLOv8 detection

The occurrence of <i>Phaeocystis globosa</i> blooms poses a potential hazard to both human society and the ecological environment, particularly concerning the safety of cooling systems in coastal nuclear power plants. However, current ecological monitoring techniques fail to dynamically detect the densities of solitary cells of <i>Phaeocystis globosa</i> prior to the blooms, thus hindering timely interventions. This study proposes a framework for harmful algae monitoring by integrating underwater microscopic imaging, image processing, and object detection. Flume experiments were conducted using <i>Phaeocystis globosa</i> as the case study for monitoring objects. The results indicate that the proposed framework exhibits favorable performance in recognizing different types of algae, particularly in distinguishing between <i>Phaeocystis globosa</i> and <i>Chlorella</i>. Despite their similar morphology observed from the underwater imaging device under dark-field illumination, the false detection rate between <i>Phaeocystis globosa</i> and <i>Chlorella</i> approaches 0% when using the YOLOv8 object detection model. Adaptive contrast enhancement (ACE) amplifies the color discrepancies among algae and eliminates the virtual focus interference, thus improving the precision of algae classification. Subsequently, dark channel prior (DCP) reduces the noise caused by image scattering and limits the missed detection. Consequently, the precision of <i>Phaeocystis globosa</i> recognition using the YOLOv8 model is increased from 74% to 91%. This study presents an effective solution for <i>in situ</i> monitoring of specific harmful algae, which has the potential to enhance the capabilities for dynamic detection and early warning of <i>Phaeocystis globosa</i> blooms.