Abstract
Accurate monitoring of microalgae cultures is crucial for long-term research on monocultures as well as for industrial scale-up processes, where maintaining culture quality directly impacts the consistency and success of the final product. Traditional microscope-based counting and sizing is labor-intensive and time-consuming, necessitating automated solutions, while lowering costs. This paper presents a web-based application, Algalytics, that integrates image processing techniques with a Convolutional Neural Network (CNN), to extract algae counts and size measurements from input images. The methodology involves image segmentation, classification into different categories which helps ensure accurate counting and sizing while avoiding the inclusion of impurities, followed by post-processing to extract measurements. The model is trained on labeled datasets, and after fine-tuning, it is deployed in a web-based application. Validation was performed on datasets featuring Porosira glacialis (a large centric diatom for CO2 sequestration) and Tisochrysis lutea (a haptophyte often used for shellfish feeding). The approach was validated on datasets with varying image quality and algae density, demonstrating the robustness of the application. Results showed high counting accuracy, while sizing was more challenging due to optical artifacts such as halo effects, which introduced measurement biases. These challenges can be mitigated by improving image focus and acquisition settings. Despite this, the system effectively captures growth trends and provides a low-cost, user-friendly, and scalable alternative to manual methods, offering consistent and efficient monitoring for both industrial operation, and small research laboratories.