Crop Discrimination Using Field Hyper Spectral Remotely Sensed Data

Accuracy of crop classification for high spatial resolution satellite imagery in the intensively cultivated lands of the Egyptian Nile delta is still low. The main objective of this research was to determine the optimal hyperspectral wavebands in the spectral range of (400 - 2500 nm) to discriminate between two winter crops (wheat and clover) and two summer crops (maize and rice). This was considered as the first step to improve crop classification through satellite imagery in the intensively cultivated areas in Egypt.

Hyperspectral ground measurements of ASD field Spec3 spectroradiometer was used to monitor the spectral reflectance profile during the period of the maximum growth stage of the four crops. 1-nm-wide was aggregated to 10-nm-wide bandwidths. After accounting for atmospheric windows and/or areas of significant noise, a total of 2150 narrow bands in 400 - 2500 nm were used in the analysis. Spectral reflectance was divided into six spectral zones: blue, green, red, near-infrared, shortwave infrared-I and shortwave infrared-II. One Way ANOVA and Tukey’s HSD post hoc analysis was performed to choose the optimal spectral zone that could be used to differentiate the different crops. Then, linear regression discrimination (LDA) was used to identify the specific optimal wavebands in the spectral zones in which each crop could be spectrally identified.

The results of Tukey’s HSD showed that blue, NIR, SWIR-1 and SWIR-2 spectral zones were more sufficient in the discrimination between wheat and clover than green and red spectral zones. At the same time, all spectral zones were quite sufficient to discriminate between rice and maize. The results of LDA showed that the wavelength zone (727:1299 nm) was the optimal to identify clover crop while three zones (350:712, 1451:1562, 1951:2349 nm) could be used to identify wheat crop. The spectral zone (730:1299 nm) was the optimal to identify maize crop while three spectral zones were the best to identify rice crop (350:713, 1451:1532, 1951:2349 nm). An average of thirty measurements for each crop was considered in the process.

These results would be used in machine learning process to improve the performance of the existing remote sensing software’s to isolate the different crops in intensive cultivated lands.

Article by Sayed M. Arafat, et al, from National Authority for Remote Sensing and Space Sciences (NARSS), Cairo, Egypt.

Full access: http://mrw.so/2c9WGm

Image by MARY LF, from Flickr-cc.