土地覆盖变化检测中不同相对辐射归一化方法的评价

【目的】从土地覆盖变化检测性能的角度寻求一种精度较高且归一化过程相对简单、客观和自动化的辐射归一化方法。【方法】借助2013年7月和2015年9月的南京地区Landsat 8 OLI图像并以2013年影像为基准,执行基于伪不变特征(pseudo-invariant features,PIF)、时不变点群(temporally invariant cluster,TIC)、全景影像回归(wall-to-wall regression,WWR)和加权不变点(weighted invariant pixels,WIP)4种方法,分波段对2015年的多光谱图像进行辐射归一化处理。在此基础上利用变化向量分析技术(change vector analysis,CVA)进行土地覆盖变化信息提取,并借助高空间分辨率Google Maps对提取到的变化信息进行空间一致性验证。【结果】使用PIF、TIC、WWR和WIP方法对图像归一化后,利用变化向量分析提取的土地覆盖变化信息与目视解译得出的土地覆盖变化信息的空间一致性分别为79.63%、81.75%、72.72%和82.59%;依不同变化地类给出变化检测阈值,算法的空间一致性会进一步提高。【结论】考虑到算法的客观性、自动化程度和变化检测精度,在4种方法中,TIC法和WIP法更适合工程化的图像辐射归一化操作。

Performance assessment of different relative radiometric normalization approaches applied to land cover change detection

Objective] Seek a relatively simple,automated and objective normalization method with a high accuracy in land cover change detection.Method] Considering the Landsat OLI 2013 (July) images as the reference,four relative radiometric normalization approaches including the pseudo-invariant features (PIF),the temporally invariant cluster (TIC),the wall-to-wall regression (WWR),and the weighted invariant pixels (WIP) were used to normalize the Landsat OLI 2015 (September) image.Based on normalization,the change vector analysis (CVA) method was used to detect land cover change information,and these detected land cover changes were validated by visually interpreting high-resolution Google images covering the study area to calculate a spatial agreement index.Result] Results indicated that the spatial agreement measures for PIF,TIC,WWR and WIP methods were 79.63％,81.75％,72.72％ and 82.59％,respectively,and the agreement measure would further escalate if the change type dependent thresholds were properly specified.Conclusion] With respect to objectivity,automation degree and change detection accuracy of the normalization methods,the TIC and WIP approaches are considered to be suitable for engineering oriented image normalization operations.