Related Experiment Video
Updated: Sep 25, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.6K
Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning
Fei Li1,2, Jingya Bai1,2, Mengyun Zhang3,4
1College of Mechanical and Electrical Engineering, Shihezi University, Shihezi, 832000, Xinjiang, People's Republic of China.
Plant Methods
|April 28, 2022
Summary
Accurate cotton yield estimation in China is now possible using low-altitude drone imagery and a specialized deep learning model. This method achieves a low average error of 6.2% for densely planted fields.
Area of Science:
- Agricultural remote sensing
- Precision agriculture
- Crop yield estimation
Background:
- Unique cotton planting patterns in Xinjiang, China, with dense, alternating rows, pose challenges for traditional remote sensing yield estimation due to occluded and overlapped branches.
- Accurate estimation of cotton yield in these complex fields is crucial for agricultural management and food security.
Purpose of the Study:
- To develop and validate a novel method for accurately estimating densely planted cotton yield using unmanned aerial vehicle (UAV) imaging and deep convolutional neural networks (DCNN).
- To address the limitations of existing remote sensing techniques in complex planting environments.
Main Methods:
- Acquisition of high-resolution cotton field images using UAVs at a low altitude (5 m).
- Development of a modified DCNN model (CD-SegNet) featuring an reorganized encoder-decoder architecture and dilated convolutions for pixel-level segmentation of cotton bolls.
- Application of linear regression to correlate the ratio of cotton boll pixels with actual cotton yield.
- Validation of the model's performance against manually harvested and weighed cotton yields.
Main Results:
- The CD-SegNet model demonstrated superior performance compared to SegNet, Support Vector Machine (SVM), and Random Forest (RF) models in cotton boll segmentation and yield estimation.
- The developed method achieved a low average error of 6.2% in estimating cotton yields across four different fields.
- Pixel-level segmentation of cotton bolls proved effective in overcoming occlusion and overlap issues inherent in dense planting patterns.
Conclusions:
- Low-altitude UAV imaging combined with advanced DCNN models provides a feasible and accurate approach for estimating densely planted cotton yields.
- This study offers a valuable methodological reference for improving cotton yield estimation in China and similar agricultural settings.
- The CD-SegNet model presents a significant advancement in applying deep learning for crop yield monitoring in challenging field conditions.
