AI-based image analysis and detection of microalgae

Microalgae, as small aquatic organisms, significantly impact the stability of aquatic ecosystems through their community dynamics. Abnormal proliferation of microalgae usually leads to water quality deterioration and ecological imbalances. Currently, bright-field microscopy method has been widely applied into microalgae detection, providing raw images for species recognition and growth status analysis. However, this method is usually time-costing and inaccurate in complicated samples, which is thereby insufficient for the demands of modern aquatic ecological monitoring and the microalgal cultivation industry. Here, we have applied artificial intelligence (AI) into microalgae detection and analysis to achieve higher accuracy and efficiency. The YOLOv5 model was used as a deep learning-based artificial intelligence object detection algorithm Microalgae were cultured for raw image acquisition under microscope in a bright-field mode. Then the images were preprocessed for YOLOv5 model training. After multiple rounds of parameter tuning to obtain suitable training parameters, the resulting model achieved a mean Average Precision (mAP) of 0.78 at an Intersection over Union (IoU) threshold of 0.5. Meanwhile, the average confidence, calculated with the confidence threshold set to 0.25, reached 0.42. The enhanced YOLOv5 model achieved an satisfactory balance between detection accuracy and coverage, resulting in improved efficacy for microalgae image processing. Briefly, this AI-Based image analysis method has great application potential in species recognition and quantification of microalgae, providing a useful evaluation tool for ecological water quality assessment and microalgae cultivation management.