AI-Augmented Pollen Recognition in Optical and Holographic Microscopy for Veterinary Imaging
Analysis
This research paper explores the use of AI, specifically YOLOv8s and MobileNetV3L, to automate pollen recognition in veterinary imaging using both optical and digital in-line holographic microscopy (DIHM). The study highlights the challenges of pollen recognition in DIHM images due to noise and artifacts, resulting in significantly lower performance compared to optical microscopy. The authors then investigate the use of a Wasserstein GAN with spectral normalization (WGAN-SN) to generate synthetic DIHM images to augment the training data. While the GAN-based augmentation shows some improvement in object detection, the performance gap between optical and DIHM imaging remains substantial. The research demonstrates a promising approach to improving automated DIHM workflows, but further work is needed to achieve practical levels of accuracy.
Key Takeaways
- •AI can be used to automate pollen recognition in veterinary imaging.
- •DIHM images present challenges for pollen recognition due to noise and artifacts.
- •GAN-based augmentation can improve object detection in DIHM images, but further improvements are needed.
“Mixing real-world and synthetic data at the 1.0 : 1.5 ratio for DIHM images improves object detection up to 15.4%.”