Objective In order to know the dynamic changes of forest resources, detect the infringement of forest land promptly, quickly and accurately, and solve the application bottlenecks of mainstream remote sensing change detection methods such as high requirements of consistency for data source and data time, too much manual interventions and complicated process, this research used a multi-scale object-level segmentation and change extraction method based on high spatial resolution time-series images to merge and simplify the two processes of classification and detection in mainstream methods.
Method Taking Baishui County in Shaanxi Province of northwestern China as the research area, using GF-1 and ZY-3 satellite data sources, the two remote sensing image bands were split and reorganized to form the time-series image, and the time-series image was segmented in a multi-scale object-oriented manner. The spectral change value of the segmentation result was statistically sampled to determine the critical point and establish the extraction threshold, and then the NDVI change value was used to optimize the result.
Result Taking the results of manual visual interpretation as a reference, the detection accuracy of this method was 86.2%. Among the classes that successfully detected invading forests, classes with good or almost consistent shapes accounted for 48.8%.
Conclusion This method can realize the rapid detection of invaded forest classes, and can meet the practical application needs of large-scale, multi-temporal, and mixed image data source forest resource monitoring in terms of detection efficiency, accuracy and adaptability.
DownLoad: