Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning
Abstract Background China has a unique cotton planting pattern. Cotton is densely planted in alternating wide and narrow rows to increase yield in Xinjiang, China, causing the difficulty in the accurate estimation of cotton yield using remote sensing in such field with branches occluded and overlapp...
Main Authors: | , , , |
---|---|
Format: | Article |
Language: | English |
Published: |
BMC
2022-04-01
|
Series: | Plant Methods |
Subjects: | |
Online Access: | https://doi.org/10.1186/s13007-022-00881-3 |
_version_ | 1811289390323859456 |
---|---|
author | Fei Li Jingya Bai Mengyun Zhang Ruoyu Zhang |
author_facet | Fei Li Jingya Bai Mengyun Zhang Ruoyu Zhang |
author_sort | Fei Li |
collection | DOAJ |
description | Abstract Background China has a unique cotton planting pattern. Cotton is densely planted in alternating wide and narrow rows to increase yield in Xinjiang, China, causing the difficulty in the accurate estimation of cotton yield using remote sensing in such field with branches occluded and overlapped. Results In this study, unmanned aerial vehicle (UAV) imaging and deep convolutional neural networks (DCNN) were used to estimate densely planted cotton yield. Images of cotton fields were acquired by the UAV at an altitude of 5 m. Cotton bolls were manually harvested and weighed afterwards. Then, a modified DCNN model (CD-SegNet) was constructed for pixel-level segmentation of cotton boll images by reorganizing the encoder-decoder and adding dilated convolutions. Besides, linear regression analysis was employed to build up the relationship between cotton boll pixels ratio and cotton yield. Finally, the estimated yield for four cotton fields were verified by weighing harvested cotton. The results showed that CD-SegNet outperformed the other tested models, including SegNet, support vector machine (SVM), and random forest (RF). The average error in yield estimates of the cotton fields was as low as 6.2%. Conclusions Overall, the estimation of densely planted cotton yields based on low-altitude UAV imaging is feasible. This study provides a methodological reference for cotton yield estimation in China. |
first_indexed | 2024-04-13T03:55:10Z |
format | Article |
id | doaj.art-10880716dcd74399b54a6d86397b98e2 |
institution | Directory Open Access Journal |
issn | 1746-4811 |
language | English |
last_indexed | 2024-04-13T03:55:10Z |
publishDate | 2022-04-01 |
publisher | BMC |
record_format | Article |
series | Plant Methods |
spelling | doaj.art-10880716dcd74399b54a6d86397b98e22022-12-22T03:03:40ZengBMCPlant Methods1746-48112022-04-0118111110.1186/s13007-022-00881-3Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learningFei Li0Jingya Bai1Mengyun Zhang2Ruoyu Zhang3College of Mechanical and Electrical Engineering, Shihezi UniversityCollege of Mechanical and Electrical Engineering, Shihezi UniversityCollege of Mechanical and Electrical Engineering, Shihezi UniversityCollege of Mechanical and Electrical Engineering, Shihezi UniversityAbstract Background China has a unique cotton planting pattern. Cotton is densely planted in alternating wide and narrow rows to increase yield in Xinjiang, China, causing the difficulty in the accurate estimation of cotton yield using remote sensing in such field with branches occluded and overlapped. Results In this study, unmanned aerial vehicle (UAV) imaging and deep convolutional neural networks (DCNN) were used to estimate densely planted cotton yield. Images of cotton fields were acquired by the UAV at an altitude of 5 m. Cotton bolls were manually harvested and weighed afterwards. Then, a modified DCNN model (CD-SegNet) was constructed for pixel-level segmentation of cotton boll images by reorganizing the encoder-decoder and adding dilated convolutions. Besides, linear regression analysis was employed to build up the relationship between cotton boll pixels ratio and cotton yield. Finally, the estimated yield for four cotton fields were verified by weighing harvested cotton. The results showed that CD-SegNet outperformed the other tested models, including SegNet, support vector machine (SVM), and random forest (RF). The average error in yield estimates of the cotton fields was as low as 6.2%. Conclusions Overall, the estimation of densely planted cotton yields based on low-altitude UAV imaging is feasible. This study provides a methodological reference for cotton yield estimation in China.https://doi.org/10.1186/s13007-022-00881-3Yield estimationUnmanned aerial vehicleSegNetDensely planted cotton |
spellingShingle | Fei Li Jingya Bai Mengyun Zhang Ruoyu Zhang Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning Plant Methods Yield estimation Unmanned aerial vehicle SegNet Densely planted cotton |
title | Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning |
title_full | Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning |
title_fullStr | Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning |
title_full_unstemmed | Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning |
title_short | Yield estimation of high-density cotton fields using low-altitude UAV imaging and deep learning |
title_sort | yield estimation of high density cotton fields using low altitude uav imaging and deep learning |
topic | Yield estimation Unmanned aerial vehicle SegNet Densely planted cotton |
url | https://doi.org/10.1186/s13007-022-00881-3 |
work_keys_str_mv | AT feili yieldestimationofhighdensitycottonfieldsusinglowaltitudeuavimaginganddeeplearning AT jingyabai yieldestimationofhighdensitycottonfieldsusinglowaltitudeuavimaginganddeeplearning AT mengyunzhang yieldestimationofhighdensitycottonfieldsusinglowaltitudeuavimaginganddeeplearning AT ruoyuzhang yieldestimationofhighdensitycottonfieldsusinglowaltitudeuavimaginganddeeplearning |