Researchers develop deep learning system to track liquid breakup into droplets and ligaments from high-speed video

Two-stage deep learning framework uses Faster R-CNN with ResNet-50 backbone to detect and classify ligaments and droplets in shadowgraphy images

Addresses key limitation of conventional tracking systems by modeling one-to-many fragmentation events instead of strict one-to-one associations

Captures ligament deformation, fragmentation, and parent-child lineage relationships during liquid sheet disintegration

Tested on high-speed recordings of impinging Carbopol gel jets to improve spray analysis quantification