Accurate weed mapping and prescription map generation based on fully convolutional networks using UAV imagery

Authors: HUANG, HUASHENG - South China Agricultural University; DENG, JIZHONG - South China Agricultural University; LAN, YUBIN - South China Agricultural University; YANG, AQING - South China Agricultural University; DENG, XIAOLING - South China Agricultural University; WEN, SHENG - South China Agricultural University; Zhang, Huihui; ZHANG, YALI - South China Agricultural University

Submitted to: Sensors
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/28/2018
Publication Date: 10/1/2018
Citation: Huang, H., Deng, J., Lan, Y., Yang, A., Deng, X., Wen, S., Zhang, H., Zhang, Y. 2018. Accurate weed mapping and prescription map generation based on fully convolutional networks using UAV imagery. Sensors. 18(10):3299. https://doi.org/10.3390/s18103299.
DOI: https://doi.org/10.3390/s18103299

Interpretive Summary: Chemical control is necessary to ensure the crop yield. However, excessive use of chemicals has increased the social concerns on substantial development of modern agriculture. Site specific weed management (SSWM) recommends chemical reduction in application and adequate herbicide according to the weed coverage, which may reduce the use of herbicide while enhancing the chemical effects. In the context of SSWM, the prescription map must be generated for the accurate spraying. In this study, high-resolution imagery taken by an unmanned aerial vehicle over a rice field. Different methods were evaluated to generate the weed coverage map for the whole field. All the experimental results demonstrated that the proposed method has potential to produce accurate weed coverage and prescription maps and reduce herbicide application.

Technical Abstract: Chemical control is necessary to ensure the crop yield. However, excessive use of chemicals has increased the social concerns on substantial development of modern agriculture. Site specific weed management (SSWM) recommends chemical reduction in application and adequate herbicide according to the weed coverage, which may reduce the use of herbicide while enhancing the chemical effects. In the context of SSWM, the prescription map must be generated for the accurate spraying. In this paper, the UAV imagery over a rice field were captured at high resolution and overlapping. Different workflows were evaluated to generate the weed cover map for the whole field. Fully Convolutional Network (FCN) was applied for pixel-level classification. Theoretical analysis and practical evaluation were carried out to seek for architecture improvement and performance boost. A chessboard segmentation process was used to build the grid framework of the prescription map. Experimental results showed that the overall accuracy and mean intersection over union (mean IU) for the weed mapping using FCN-4s were 0.9196 and 0.8473, and the total time (including data collection and data processing) required to generate the weed cover map for the entire field (50×60 m) was less than half an hour. Different weed thresholds (0.00-0.25, with an interval of 0.05) were used for the prescription map generation. High accuracies (above 0.94) were observed for all threshold values, and relevant herbicide saving ranged from 58.3% to 70.8%. All the experimental results demonstrated that the method used in this work has potential to produce accurate weed cover map and prescription map in SSWM applications.