Analysis of Spectral Change of Cotton During Growth Period Based on Endmember Extraction of UAV Hyperspectral Image

【Objective】 Studying the trend of reflection spectrum in the time series of cotton growth cycle is of great significance for cotton growth monitoring. 【Method】 In order to obtain the spectrum curve of cotton at different growth stages and its changing rule. In this paper, a multi-rotor unmanned aerial vehicle(UAV) M600 PRO carrying a Rikola hyperspectral imager was used to collect a total of five periods of cotton hyperspectral data from flowering period to later period. Moguhu village, Shawan county, Tacheng district, Xinjiang (85°51'49.44' E, 44°25'26.61' N) was selected as the study area. We collected hyperspectral data between 12 and 14 noon. UAV flied 3 sorties per period, flying at altitudes of 60, 80 and 100 m respectively.And the flight time of each sortie was about 13 minutes. Meanwhile, SR-3500 portable spectrometer was used to collect cotton canopy spectral data on the ground for verification.The acquired hyperspectral data of UAV were preprocessed by lens halo correction, dark current correction, band registration, image Mosaic, and reflectivity correction, etc. Finally we got usable hyperspectral images.The pixel purity index(PPI) endmember extraction algorithm and the N-FINDR endmember extraction algorithm were used to extract the cotton spectrum curve from the UAV hyperspectral images. Then, the spectral curve collected by the SR-3500 spectrometer on the ground was taken as the standard. Spectral angle mapping(SAM) was used as the evaluation index to verify the endmember extraction effect. 【Result】 The results showed that The spectral angles of the spectral curves which extracted by the N-FINDR algorithm in the hyperspectral images at 60, 80 and 100 m altitude were 0.065,8, 0.065,9 and 0.067,7. It was closer to the SR-3500 spectrometer data than the results of the PPI algorithm. N-FINDR algorithm could better extract small sample endmember. The spatial resolution change caused by the variation of altitude below 100m has little impact on the reflection spectrum, with the result difference was less than 2%. Analyzeing to the spectrum curve of cotton in different growth stages, the absorption valley and red edge reached the peak value on July 18.The maximum values of NDVI and RVI reached 0.841,7 and 11.630,5 on July 11, while the maximum values of EVI, DVI and OSAVI reached 0.818,9, 0.501,3 and 0.501,2 on July 18. 【Conclusion】 The N-FINDR endmember extraction algorithm can extract the cotton spectral curve from the cotton hyperspectral image.100 m is the better altitude of UAV data acquisition.The photosynthesis of cotton at its peak in july. The strong absorption of red light and the strong reflection of near infrared wave band are the most obvious during peak flowering period and bell period. This paper can provide reference for crop growth monitoring based on UAV remote sensing.