Related Experiment Video
Updated: Jan 17, 2026
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.8K
MOSSNet: multiscale and oriented sorghum spike detection and counting in UAV images
Jianqing Zhao1, Zhiyin Jiao2,3, Jinping Wang2,3
1Key Laboratory for Climate Risk and Urban-Rural Smart Governance, School of Geography, Jiangsu Second Normal University, Nanjing, China.
Frontiers in Plant Science
|September 15, 2025
Summary
Accurate sorghum spike detection is crucial for crop monitoring and yield prediction. A new model, MOSSNet, effectively counts sorghum spikes in UAV images, outperforming existing methods in complex field conditions.
Area of Science:
- Agricultural Science
- Computer Vision
- Artificial Intelligence
Background:
- Accurate sorghum spike detection is vital for crop monitoring, yield prediction, and food security.
- Deep learning models offer improved accuracy but struggle with dense, variable sorghum spike features in UAV imagery.
- Challenges include dense distribution, varied sizes, and complex backgrounds in aerial images.
Purpose of the Study:
- To develop a robust model for accurate sorghum spike detection and counting in UAV images.
- To enhance feature extraction for small and variably oriented sorghum spikes.
- To improve the efficiency and accuracy of sorghum spike analysis in agricultural settings.
Main Methods:
- Proposed MOSSNet (Multiscale and Oriented Sorghum Spike detection) model for UAV images.
- Integrated Deformable Convolution Spatial Attention (DCSA) module for enhanced feature capture.
- Employed Circular Smooth Labels (CSL) for morphological representation and Wise IoU loss for localization.
Main Results:
- MOSSNet achieved 90.3% mAP in field conditions for sorghum spike counting.
- Demonstrated superior performance in predicting spike orientation (RMSEa: 14.6, MAEa: 12.5).
- Outperformed general object detection algorithms in counting accuracy (RMSE: 9.3, MAE: 8.1).
Conclusions:
- MOSSNet effectively handles dense, occluded, and complex background scenes in sorghum spike detection.
- The model shows robustness and generalizability for agricultural applications.
- Future work includes exploring MOSSNet across different growth stages and developing real-time detection.
