Related Experiment Video
Updated: Jun 25, 2026
08:47
Computer Vision-Based Biomass Estimation for Invasive Plants
Published on: February 9, 2024
Partially occluded weed classification using vision transformers and convolutional neural networks for precision
Euan Hall1, Emmanuel Junior Zuza2, Karen Rial Lovera1
1School of Agricultural Science and Practice, Royal Agricultural University, Cirencester, GL7 6JS, UK.
Scientific Reports
|June 23, 2026
Summary
Vision Transformers (ViTs) demonstrate superior resilience to occlusion in automated weed detection compared to Convolutional Neural Networks (CNNs). This finding is crucial for improving AI-driven precision agriculture and reducing herbicide use.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Artificial Intelligence
Background:
- Automated weed detection is key for site-specific herbicide application, reducing environmental impact.
- Occlusion poses a significant challenge to the precision of automated weed identification in field deployments.
Purpose of the Study:
- To compare the performance of Vision Transformers (ViT-B16, PvTv2) and Convolutional Neural Networks (EfficientNet-B0, ResNet-50) for accurate weed detection under varying occlusion levels.
- To evaluate the occlusion resilience of different AI architectures in the context of automated weed detection.
Main Methods:
- Controlled synthetic occlusion levels (0%, 25%, 50%) were applied to images for testing.
- Performance comparison of ViT-B16, PvTv2, EfficientNet-B0, and ResNet-50 models.
- Multivariable regression analysis to identify key drivers of testing accuracy.
Main Results:
- ViT-B16 exhibited superior occlusion resilience, with accuracy increasing from 80% to 86% under 50% occlusion.
- Accuracy of PvTv2, EfficientNet-B0, and ResNet-50 dropped significantly (45-76%) under 50% occlusion.
- Architecture type was a dominant accuracy driver (p≤0.001), with ViTs outperforming CNNs by an average of 14.56% points.
Conclusions:
- Occlusion resilience varies significantly across AI architectures, with attention-based designs like ViTs showing better performance.
- Hybrid architectures balancing ViT's global context and CNN's efficiency are a promising direction for real-time weed detection.
- Improved weed detection accuracy supports precise herbicide application, reduced chemical inputs, and sustainable crop protection.
