Development of systematic algae strain and biomass detection method and database with automation engineering

Microalgae are an emerging renewable energy source due to their rapid growth and high lipid content, making them suitable for biofuel production. However, the economic challenges associated with microalgae-based biofuels necessitate innovative technologies to enhance identification, prediction, and classification efficiency, paving the way for cost-effective biorefineries. This study aimed to establish a precise, real-time, and cost-effective system for microalgae biomolecule quantification and identification. By leveraging machine learning (ML) and deep learning (DL) techniques, we successfully digitalised the prediction of C-phycocyanin (CPC) content from Spirulina platensis images. Our findings revealed that support vector machines (SVM) and artificial neural networks (ANN) achieved high accuracy when incorporating additional parameters like 'Abs' and 'Day', outperforming convolutional neural networks (CNNs). To simulate real-world scenarios, we analysed the impact of various input parameters under different lighting conditions and devices. The XGBoost meta-regressor demonstrated superior performance, offering enhanced stability and generalisation, particularly in challenging light-disturbed environments. Furthermore, our investigation into microalgae classification across three species (Chlorella vulgaris FSP-E, Chlamydomonas reinhardtii, and Spirulina platensis) showcased remarkable results. By optimising image pre-processing techniques, k-nearest neighbours (k-NN) and SVM achieved accuracies of 96.93% and 97.63%, respectively. The Azure Custom Vision model further excelled, reaching an impressive 97.86% accuracy. To advance real-time monitoring, we deployed the YOLOv8 model for microalgae detection and instance segmentation. The YOLOv8-n box detection model achieved high precision and recall, while its box instance segmentation outperformed alternatives with exceptional accuracy and reliability. Finally, to address the scarcity of high-quality microalgae datasets, we implemented generative AI techniques. Models like FastGAN, VQVAE, and DDIM effectively synthesised realistic images of Chlorella vulgaris FSP-E, Chlamydomonas reinhardtii, and Spirulina platensis significantly improving classification capabilities and dataset diversity.