CropNeRF: A Neural Radiance Field-Based Framework for Crop Counting
Md Ahmed Al Muzaddid, William J. Beksi
AI summary
Problem
Counting crops in outdoor fields is hindered by partial occlusions, clustered instances, and perspective ambiguity, which degrade the accuracy of traditional 2D image-based segmentation methods.
Approach
The framework reconstructs a 3D semantic point cloud from multi-view images using a NeRF, then partitions and merges point clusters based on cross-view visibility and mask consistency scores to achieve exact 3D instance segmentation.
Key results
- Accurate counting across cotton, apple, and pear datasets despite varying crop dimensions
- Robust handling of occlusions and 2D annotation errors via visibility and consistency scores
- Elimination of crop-specific parameter tuning and manual recalibration
- Release of a new public infield cotton plant dataset for 3D counting research
Why it matters
Provides a scalable, crop-agnostic solution for precise yield estimation and resource optimization in precision agriculture and robotic harvesting.
Abstract
Rigorous crop counting is crucial for effective agricultural management and informed intervention strategies. However, in outdoor field environments, partial occlusions com- bined with inherent ambiguity in distinguishing clustered crops from individual viewpoints poses an immense challenge for image-based segmentation methods. To address these problems, we introduce a novel crop counting framework designed for exact enumeration via 3D instance segmentation. Our approach utilizes 2D images captured from multiple viewpoints and associates independent instance masks for neural radiance field (NeRF) view synthesis. We introduce crop visibility and mask consistency scores, which are incorporated alongside 3D information from a NeRF model. This results in an effective segmentation of crop instances in 3D and highly-accurate crop counts. Furthermore, our method eliminates the dependence on crop-specific parameter tuning. We validate our framework on three agricultural datasets consisting of cotton bolls, apples, and pears, and demonstrate consistent counting performance despite major variations in crop color, shape, and size. A com- parative analysis against the state of the art highlights superior performance on crop counting tasks. Lastly, we contribute a cotton plant dataset to advance further research on this topic.