A Deep Learning Approach for Weed Detection in Lettuce Crops Using Multispectral Images
Problem
Farmers and experts have long struggled to estimate the amount and location of weeds in lettuce crops, which is key to crop productivity.
Why it matters
Weed management is one of the most important aspects of crop productivity, per the abstract.
What the AI did
Three deep/machine learning methods (SVM with HOG, YOLOv3, Mask R-CNN) detected and segmented weeds from multispectral images, aided by an NDVI background subtractor, and were compared to expert visual estimation.
Results
F1-scores of 88% and 94% for the machine and deep learning methods (abstract truncated before the third value)
Impact
The AI methods matched or approached expert visual weed estimation on the chosen metrics.
Record
| Venue | AgriEngineering |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Kavir Osorio, Andrés Puerto, Cesar Augusto Pedraza, David Alejandro Jamaica-Tenjo, Leonardo Rodríguez |
| Citations | 236 (OpenAlex, at import time) |
| References | 51 |
| Model or method | SVM with HOG, YOLOv3, Mask R-CNN, NDVI index |
| Organization | not stated in source |
| Primary source | https://doi.org/10.3390/agriengineering2030032 |
| Open access | gold |
| Keywords | weed detection, deep learning, YOLOv3, Mask R-CNN, multispectral imaging, lettuce |
Importance rating
incremental / niche
Weed management is called important for crop productivity, but scope is limited to lettuce crops and no scale is stated; incremental precision-agriculture advance.