Evaluation of spatial resolution on crop disease detection based on multiscale images and category variance ratio

Authors: ZHAO, HENGQIAN - China University Of Mining And Technology; YANG, YIFENG - China University Of Mining And Technology; Yang, Chenghai; SONG, RUI - China University Of Mining And Technology; GUO, WEI - Henan Agricultural University

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/26/2023
Publication Date: 3/8/2023
Citation: Zhao, H., Yang, Y., Yang, C., Song, R., Guo, W. 2023. Evaluation of spatial resolution on crop disease detection based on multiscale images and category variance ratio. Computers and Electronics in Agriculture. https://doi.org/10.1016/j.compag.2023.107743.
DOI: https://doi.org/10.1016/j.compag.2023.107743

Interpretive Summary: Crop diseases are a main cause for crop losses in agricultural production worldwide. Remote sensing has long been used for crop disease detection, but it is necessary to select a suitable imaging platform and spatial resolution for practical applications. In this study, the category variance ratio (CVR) method was proposed to characterize the separability of healthy and cotton root rot-infested cotton areas on images from three different platforms, including an unmanned aerial vehicle (UAV), a manned aircraft, and the Sentinel-2 satellite. Based on the CVR separability index, the manned aircraft-based imagery had the best separation between healthy and root rot areas among the three types of images. Further analysis of the upscaled UAV images at different spatial resolutions showed that the 4-m resolution in the reconstructed image was the optimal resolution for cotton root rot identification. The results from this study demonstrate that the CVR method was effective for selecting the most appropriate spatial resolution and image type to extract crop disease information.

Technical Abstract: Crop diseases are a main cause for crop losses in agricultural production worldwide. Remote sensing has long been used for crop disease detection, but it is necessary to select a suitable imaging platform and spatial resolution for practical applications. In this study, the category variance ratio (CVR) method was proposed to characterize the separability of healthy and cotton root rot-infested cotton areas on images from three different platforms, including an unmanned aerial vehicle (UAV), a manned aerial vehicle (MAV) and the Sentinel-2 satellite. The visible and near-infrared bands were selected to construct several vegetation indices and the separability parameters for the three types of images were calculated and compared using multiple evaluation indicators. The comparison results showed that, based on the CVR separability index proposed in this study, the MAV image had the best separation between healthy and root rot areas among the three types of images. Through pixel clustering and resampling, the UAV images were upscaled to the resolution of the MAV images, but the upscaled UAV image was not as effective as the original MAV image. Further analysis of the upscaled UAV images at different spatial resolutions showed that the 4-m resolution in the reconstructed image was the optimal resolution for cotton root rot identification. Image classification and accuracy assessment results showed that the resampled 4-m UAV and MAV images had better classification accuracy than the original images and that the upscaled 4-m MAV image achieved the best accuracy among the three types of images. The results from this study demonstrate that the CVR method was effective for selecting the most appropriate spatial resolution and image type to extract crop disease information.