Related Experiment Video
Updated: Jul 16, 2025
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.3K
WheatSpikeNet: an improved wheat spike segmentation model for accurate estimation from field imaging
M A Batin1, Muhaiminul Islam1, Md Mehedi Hasan1
1Department of Robotics and Mechatronics Engineering, University of Dhaka, Dhaka, Bangladesh.
Frontiers in Plant Science
|September 11, 2023
Summary
Accurate wheat spike counting is crucial for phenotyping and yield estimation. This study introduces WheatSpikeNet, an enhanced instance segmentation model that achieves state-of-the-art performance in detecting and segmenting wheat spikes from field images.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Breeding
Background:
- Plant phenotyping is vital for identifying desirable crop traits and assessing environmental impacts.
- Wheat spike density is a key agronomic factor, but accurate counting in diverse field conditions remains challenging.
Purpose of the Study:
- To develop an efficient and accurate method for segmenting and counting wheat spikes in field imagery.
- To introduce a curated dataset (SPIKE-segm) and an optimized instance segmentation model (WheatSpikeNet).
Main Methods:
- Utilized the Cascade Mask RCNN architecture with enhancements and hyperparameter tuning.
- Employed ResNet50 with Deformable Convolution Network (DCN) as the backbone, Generic RoI Extractor (GRoIE), and Side Aware Boundary Localization (SABL).
- Conducted comprehensive ablation analysis to determine the most efficient model components.
Main Results:
- WheatSpikeNet achieved superior performance on the SPIKE dataset with bbox and mask mAP scores of 0.9303 and 0.9416, respectively.
- Demonstrated state-of-the-art detection and segmentation capabilities, outperforming existing methods.
- Achieved up to 0.41% mAP improvement in spike detection and 3.46% mAP improvement in segmentation tasks.
Conclusions:
- The proposed WheatSpikeNet model offers a robust solution for accurate wheat spike segmentation and counting.
- This advancement facilitates precise yield estimation in wheat breeding programs.
- The optimized model and dataset contribute to improving automated plant phenotyping techniques.
