New AI catches tiny apple leaf diseases other tools miss
Crop Improvement
If you grow apple trees, catching rust or leaf spot while it's still a few small dots on a handful of leaves, rather than after it's spread across the canopy, is the difference between a quick treatment and losing part of your harvest.
Small disease spots on apple leaves are easy for computer vision systems to miss because they're tiny and easily lost against busy backgrounds like branches and soil. This team built a smarter detection network that widens its 'field of view' around each spot and specifically practices on the hardest-to-spot examples during training. The result is a tool that found small diseases like rust and frog eye leaf spot with over 87% precision, beating a dozen other popular detection systems.
Key Findings
ALSDet achieved 65.6% mean average precision and 71.2% average recall, outperforming Cascade R-CNN, Faster R-CNN, YOLOv7, YOLOv8, and eight other detection models
For small-target diseases like rust and frog eye leaf spot, average precision exceeded 87% at an IoU threshold of 0.5
Key architecture changes include a global context module in ResNet-50 Stage2-4 and dilated convolutions in Stage2-3 bottleneck blocks to enlarge the receptive field for small disease targets
chevron_right Technical Summary
Researchers built an AI tool called ALSDet that spots small, hard-to-see disease spots on apple leaves, like rust and frog eye leaf spot, more accurately than existing detection systems, which could help orchard growers catch problems earlier.
Abstract Preview
Original paper
ALSDet: a global context-enhanced network for detecting small-target diseases on apple leaves
Accurate detection of small-target diseases on apple leaves is of great importance for optimizing orchard management and facilitating precision agriculture. Small-target disease detection remains c...
open_in_new Read full abstractAbstract copyright held by the original publisher.
Species Mentioned
Was this useful?
Want to tell us more? (optional)
Thanks for the note!
Something went wrong — please try again.
Too many submissions. Try again in an hour.
Gene editing removes 97% of celiac-triggering proteins from bread wheat
It could mean that people with celiac disease — roughly 1 in 100 worldwide — may one day safely eat bread made from real wheat, without sacrificing the taste...
An apple is the round, edible fruit of an apple tree. Fruit trees of the orchard or domestic apple, the most widely grown in the genus, are cultivated worldwide. The tree originated in Central Asia, where its wild ancestor, Malus sieversii, is still found. Apples have been grown for thousands of ...