Deep Learning-Based Weed Detection Using UAV Images: A Comparative Study
Problem
Farmers need to automatically detect weeds in crop fields from drone images so they can protect yields.
Why it matters
Weeds reduce crop yields, and reliable automated detection helps the agricultural community use technology effectively.
What the AI did
Compared deep learning segmentation models (SegNet to DeepLabV3+) with five CNN backbones on a UAV RGB dataset (CoFly-WeedDB) to identify the best weed-detection model.
Results
UNet with EfficientNetB0 backbone was the best-performing model; specific metrics not stated in source (abstract truncated)
Impact
Identifies a best-performing segmentation model for UAV-based weed detection, aiding model selection.
Record
| Venue | Drones |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Tej Bahadur Shahi, Sweekar Dahal, Chiranjibi Sitaula, Arjun Neupane, William W. Guo |
| Citations | 65 (OpenAlex, at import time) |
| References | 62 |
| Model or method | UNet with EfficientNetB0 backbone (best); SegNet, DeepLabV3+ with VGG16, ResNet50, DenseNet121, MobileNetV2 |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/drones7100624 |
| Open access | gold |
| Keywords | weed detection, UAV imagery, semantic segmentation, deep learning, precision agriculture |
Importance rating
real but narrower advance
Identifies best UAV weed-segmentation model to help protect crop yields; useful for precision agriculture, but abstract states no scale of affected area or adoption.