Abstract: [Objective] Using multipolar spaceborne SAR data, the probability density distribution characteristics of backscatter intensity ratio images were analyzed, the backscatter intensity information and image spatial context information was fused to develop a forest cover change detection method with high detection accuracy, low false alarm rate and low missing alarm rate, in order to provide technical support for the operational application of multi-polarization SAR satellite data.[Method] This study developed a forest cover change detection method that combines the "change detection method based on bi-temporal forest cover classification" (CBFC) and the "Bayesian maximum expected-Markov random field (EM-MRF) change detection method ". Firstly, based on the threshold segmentation method, the initial forest cover change map was obtained through forest/non-forest classification of bi-temporal multi-polarization SAR images. Then, Fisher feature transformation and EM-MRF classification were performed on the multi-polarization ratio image with the initial forest cover change map as training data. The results of forest cover change detection were obtained by EM-MRF iteration classification of the composite difference image converted from bi-temporal polarization (HH,HV) ratio image with Fisher feature transformation. In Xunke County, Heilongjiang Province, the effectiveness of the proposed method was evaluated based on bi-temporal ALOS PALSAR dual-polarization SAR data and the reference forest cover change map, which was obtained by visually interpretation of bi-temporal Landsat-5 images and high spatial resolution remote sensing images. And the comparative analysis was conducted between the proposed method, CBFC method and the method of combining the CBFC with EM-MRF by direct masking (CBFC-EM-MRF).[Result] The difference image obtained by Fisher feature transformation can effectively increase the contrast of the changed/non-changed category of forest cover. There are many wrongly detected small changed areas in the results of CBFC, and the false alarm rate and missed alarm rate of which are also very high. In contrast, the proposed method can improve the spatial coherence of the detection results by considering the context information of the difference image through MRF, and the false alarm rate, missed alarm rate and the accuracy were 1.58%, 11.87% and 98.36% respectively, so both the performance and the accuracy of the proposed method are better than that of CBFC and CBFC-EM-MRF.[Conclusion] The forest cover change detection method proposed in this paper has the advantages of good convergence, high reliability and demanding less user interaction, so it is of valuable reference value for the operational application of forest resource monitoring using GF-3 and the other multi-polarization SAR satellites to be launched in the future.

Key words: ALOS PALSAR, dual-polarization, forest cover, change detection, Markov random field, Bayes' rule