Faster Computer Vision Spots Defects On Bumpy Bitter Gourds

Researchers in China have built a faster computer vision system that spots surface defects on bitter gourds, whose bumpy skin makes damage hard to see. The team started with a standard real time detector and changed how it groups features, enlarges image maps and penalises poorly fitted boxes. They tested it on 1400 colour photos from fields, indoor rooms and wholesale markets, with more than 4300 marked defects. Accuracy rose from 87 point 73 percent to 90 point 35 percent, while the share of defects found rose from 84 point 91 percent to 86 point 80 percent. Speed rose from about 251 images per second to about 341 images per second on a high performance graphics card. The model made fewer mix ups between damage and pest or disease signs, though very small or faint lesions were still missed. The team now plans tests on compact hardware and production lines, plus more varieties and growing regions before commercial use.

image
advertisement
"We will never ask you to login, or register through a third-party advertisement."