干旱区生态环境知识资源中心

Change detection techniques aim to identify changes between two or more images taken at different times. In this paper, we explore the capabilities of identifying changes in an unsupervised manner between different soil types using two laboratory HySpex imaging spectroscopy sensors in the visible near infrared (VNIR) and short-wave infrared (SWIR) spectral ranges. The experiment was carried under controlled laboratory conditions with the same lighting and no atmospheric distortions. The 69 selected soil samples covered the arid and semiarid climate zones of Israel. The well-known change vector analysis technique was used to generate the difference image, and several thresholding methods were tested to generate the final binary change map. The performance capabilities of the VNIR, SWIR and combined VNIR-SWIR sensors were examined. Our study demonstrates that changes in different soil types can be identified using imaging spectroscopy sensors; the SWIR sensor generated better change detection capabilities than the VNIR sensor, and the combination of the two sensors did not outperform the SWIR sensor alone. Results showed that it is important to combine a spectral domain thresholding approach with a spatial domain thresholding approach. The benefit of combining these approaches is a low false-alarm rate with a relatively high probability of detection. Although the change experiment was conducted under almost perfect conditions without any atmospheric or lighting differences, the change detection techniques did not detect all soil type changes and changes between spectrally similar soils remain undetected. The results of this study can be further extended to other spatial scales and can provide a foundation for soil change detection using upcoming imaging spectroscopy satellite platforms that acquire spatial-spectral-temporal information. (C) 2013 Elsevier B.V. All rights reserved.