基于图像融合与多尺度分割的目标级变化检测

摘要

传统的像素级变化检测对辐射校正及阈值选择要求较高，因而在实际应用中受到诸多限制。在分析多尺度分割的基础上，提出了一种目标级的变化检测方法。分别利用粗、细尺度对各时相遥感图像的融合图像进行面向对象分割，以获取不同尺寸的目标区域，构造目标的特征进行向量分析得到差异图，并定义变化信息的强度，再利用多值逻辑理论将粗、细尺度下的检测结果进行决策级融合。实验结果表明，与传统的像素级检测方法相比，该方法受辐射差异影响小，检测精度更高，且检测结果对变化强度的衡量准确，能对应于有一定物理意义的目标变化。%The traditional pixel-level change detection algorithms have the disadvantage of high requirements for radiometric correction accuracy and threshold selection,which may limit their applications .An analysis was made to the multi-scale segmentation method,and an object-level algorithm was proposed based on it .The fused image of the remote sensing images at different time phases was segmented by using object-oriented method in coarse and fine scales respectively to capture object areas of different sizes .Discrepancy image was obtained after analyzing of characteristics vectors,and the intensity of the changed information was defined .Then,the decision-level fusion was implemented using the measurement results of the coarse and fine scales based on multi-valued logic theory .Experiments indicate that,compared with traditional pixel-level change detection algorithms,the algorithm is less influenced by radiometric discrepancy,has higher detection accuracy,and can evaluate the changed intensity accurately .