Freshwater Microscopic Algae Detection Based on Deep Neural Network with GAN-Based Augmentation for Imbalanced Algal Data

This work usesGAN-based data augmentation and machine learningto improve freshwater algal image detection, which could aid in identifyingharmful algal blooms and water quality monitoring. Identifying and quantifying algal genera in images arecrucialfor understanding their ecological impact. Algal data are often imbalanced,limiting detection model accuracy. This paper presents a novel dataaugmentation method using StyleGAN2-ADA to enhance algal image instancesegmentation. StyleGAN2-ADA generates artificial single-algal imagesto address data scarcity and imbalance. We train a Cascaded Mask R-CNNwith Swin Transformer on a combined data set of real and artificialmultigenera algal images and evaluate performance using the COCO mAPmetric. The approach improves bounding box detection performance by17.9% on all genera and 32.1% on rare genera compared with the baselinemodel. Additionally, 50% more artificial data yield significant enhancementswithout excessive artificial data use. The GAN-based augmentationtechnique shows a performance improvement in both Swin-Tiny and ResNet-50backbone models, suggesting adaptability for various machine learningmodels. The increased mAP leads to the accurate identification ofharmful algae genera, allowing for better prevention and mitigation.This method offers a superior data augmentation solution for accuratealgal instance segmentation and can benefit applications challengedby imbalanced and scarce data.