Please visit also visit the webpage page of Pollen Morphology-based Classification and Analysis for more relevant projects.
The system first detects and segments pollen grains from a pile of scans over a sample with various z-plane (focus), then finely segments the grains over cropped window and classify them. Essentially, it involves proposal detection, segmentation, and classification. For classification, one is able to adopt multi-instance mechanism on multiple crops at different slides at the same location, due to various focuses.
Run 'main003_saveData.m' to get the annotated images. The path to dataset needs to be changed accordingly.
Download models at google drive
Run 'main100_demo.m' for the demo.
Shu Kong @ UCI 06/5/2017