Cyanobacteria pose a critical challenge for freshwater management due to their ability to rapidly proliferate and produce toxins that can jeopardize human health, even at low concentrations. Therefore, effective methods to accurately classify cyanobacterial genera are essential for water quality assessment. However, existing automated classification methods often suffer from low accuracy or limitations in identifying cyanobacterial genera. To address these challenges, we propose a novel deep learning model, MobileYOLOCyano, which integrates YOLOv8 with MobileNetV4, optimizes the anchor-free detection framework, and introduces a newly designed AdaptiveChannelHead module to enhance feature extraction and genus-level classification. The MobileYOLOCyano model was evaluated on a dataset comprising nine cyanobacterial genera, which have been reported to possess the potential to produce multiple types of cyanotoxins, and achieved impressive results: 97.09 % precision, 96.82 % recall, 96.95 % F1 score, and 98.17 % mAP50. Notably, the model demonstrated significant improvements compared to previous studies in accurately classifying challenging genera (Aphanizomenon, Phormidium, Planktothrix, and Raphidiopsis), with precision and recall increasing by over 10 %. Furthermore, the model's consistently high F1 scores across a wide range of confidence thresholds demonstrates its robustness and flexibility. These results highlight the advancements made by our proposed approach and demonstrate its capability to serve as a valuable tool for water quality management and sustainable development.
Keywords: AdaptiveChannelHead; Cyanobacteria; MobileNetV4; Water quality management; YOLOv8.
Copyright © 2025. Published by Elsevier Ltd.