Forest Cover Change Detection Method Using GF-1 Multi-spectral Data

Forest is the largest terrestrial ecosystem, its change has a significant impact on global ecological environment, biological diversity and global climate. To acquire change information of forest resources accurately and timely is of great significance to global environmental change study and local forest management planning. Traditional forest resources change detection is mainly based on ground measurements, so it has some problems such as heavy workload, high cost, long cycle and low efficiency, and the survey accuracy is difficult to meet today's forest resources monitoring need. In recent years, China has launched GF-1 satellite, the 16 m spatial resolution multi-spectral data of it should be suitable for forest cover detection applications, so a forest cover change detection method were developed for GF-1 multi-spectral images by integrating some of the newly developed land cover change dection algorithms.The study site is located in Guangxi, China. Two scenes of GF-1 multi-spectral images and ground true data were acquired for evaluating the performance of the forest land cover change detection methodology.Firstly, in order to deal with the impacts of cloud/shadow in the remote sensing images on change detection algorithm, we used HOT-based method to detect the clouds and shadows. The results showed that, the method can detect most of the clouds and shadows existed in the image, and the cloud/shadow impacts on change detection were eliminated through the mask processing.Secondly, we use iteratively re-weighted multivariate change detection method to discriminated unchanged image pixels whose spectral value were used to fit one model through regression analysis for the relative radiometric normalization of the target image in order to make the target image to be in the same radimetric scale with the reference image. The results showed that this method can effectively reduces radiometric difference due to different image acquisition time and solar elevation angle, in this way the same objects in the two images can have similar reflectance level.Finally, we apply kernel minimum noise fraction(KMNF) transformation to the reference image and the normalized target image respectively to obtain the transformed features,the difference image between the first variate of the transformed features is used to automatically extracted change pixels with the threshold determined by OTSU algorithm and NDVI is utilized to determine the direction of change. Experiment shows that this method was more accurate than the traditional change vector analysis method, and the overall change detection accuracy was increased from 80.51% to 85.37%.