ObjectiveUsing high spatial resolution satellite images to capture accurate information on vegetation change is of great significance for the rational use of vegetation resources and sustainable management. Traditional pixel-based direct change detection methods are easy to cause salt-and-pepper noises and the results of object-oriented classification methods depend heavily on the classification accuracy. After investigating the advantages and weaknesses of the existing algorithms for vegetation change analysis, the major objective of this study was to develop a relatively objective algorithm for vegetation change detection by high spatial resolution remote sensing data, and to verify its effectiveness.
MethodAn object-oriented multi-index integrated change analysis (MIICA) algorithm was proposed in the analysis based on a existing MIICA. First, bi-temporal cross-sensor high spatial resolution images were segmented uniformly with the optimal segmentation parameters, which were determined by examining the precision (P) and recall (R) indices, followed by the extraction of feature parameters of the segmented objects. Then the appropriate thresholds objectively determined by ROC (receiver operating characteristic) curves were integrated to derive vegetation change positions and directions (vegetation gain or loss) finally.
ResultResults showed that compared with the pixel-based MIICA and the object-oriented classification method, the producer’s accuracy of our method was higher than that of the pixel-based MIICA, meanwhile the user’s accuracy was higher than that of the object-oriented classification method. And our method had higher overall accuracy and Kappa coefficient, estimated at 0.880 and 0.805, respectively. The detection results of our method could better reflect the positions and shapes of the vegetation change areas, with some subtle vegetation changes clearly detected.
ConclusionObject-oriented MIICA can improve the shortcomings of pixel-based MIICA and object-oriented classification method, and improve the detection accuracy. The proposed method is essential for the analysis of vegetation changes, the reasonable utilization and sustainable management of vegetation resources in forest parks or natural reserves, where heavy anthropogenic disturbances frequently exist.