<![CDATA[Microalgae are unicellular microscopic organisms that live in various water. Microalgae such as Spirulina platensis and Chlorella vulgaris are grown due to their potential as bioenergy source. During cultivation, typically, hemocytometers are used to manually count the cells and that is time-consuming and prone to human error. This research aims to develop microalgae detection model based on microscopic images and Convolutional Neural Network using YOLOv8 architecture. The methodology includes sample preparation (dilution and optical density measurement), best density determination, image acquisition, annotation, creation of datasets, YOLOv8 model training, and model performance evaluation. Best density determines good microscopic images. Image acquisition was done using binocular microscope and acquired 560 images which were then annotated. The YOLOv8n, YOLOv8s, and YOLOv8m models were trained using default hyperparameters on Google Collaboratory to determine the augmentation effect on model accuracy. Model performance evaluation was done on selected YOLOv8 models. The results showed the augmentation (crop, brightness, blur) get the highest mAP train and test on YOLOv8m model, which are 0.945 and 0.913. The YOLOv8m model was retrained with various hyperparameters and it was found that the best configuration was SGD optimizer, epoch 50, and learning rate 0.01 with mAP train and test are 0.934 and 0.925. However, 29 epochs yielding a better model with accuracy of 0.8535, minimising overfitting and resource wastage. This research can facilitate the more efficient and automatic counting for microalgae-related research and industry.]]>
