Potential Evaluation of High Spatial Resolution Multi-Spectral Images Based on Unmanned Aerial Vehicle in Accurate Recognition of Crop Types

Abstract

The accurate acquisition of farmland planting information is the basis of precision agriculture. Collecting remote sensing data via unmanned aerial vehicle (UAV) is a convenient method to obtain precision agricultural information because of the high spatiotemporal resolution and flexibility. A quadrotor UAV equipped with a SEQUOIA sensor (one multi-spectral sensor and one RGB lens) was operated over the Jingyuetan agricultural area with five land cover types on September 4, 2017, to investigate the equipment’s feasibility for crop identification. To evaluate the effects of different data and classification methods on the accuracy of crop type classification, three combinations were tested: MDC + Four (Mahalanobis distance classifiers based on four-band reflectance), MDC + VIs (Mahalanobis distance classifiers based on Vegetation Indices) and MLC + VIs (maximum likelihood classifiers based on Vegetation Indices). The accuracy of the different classification methods was 83.06% (MDC + Four), 89.17% (MDC + VIs) and 92.60% (MLC + VIs). The MLC + VIs scheme was the most accurate, as it could partially overcome the influence of shadow and flattened. Since the reflectivity of different bands varied, all kinds of objects on the ground could be distinguished. This result revealed that multi-spectral UAV technology has the potential to identify crop type at the sub-meter spatial resolution, with the MLC based on the VIs.