基于多尺度分割和缨帽变换的高原山区森林覆盖变化检测

Forest cover change detection based on multi-scale segmentation and tasseled cap transformation over plateau area

【目的】基于影像多尺度分割和缨帽变换,构建适用于高原复杂山区森林覆盖变化的检测方法。【方法】应用Landsat 5 TM和Landsat 8 OLI多光谱影像,以滇中东高原过渡山区地带为研究区,基于多尺度分割,利用缨帽变换(TCT)和主成分分析(PCA),重构面向对象多特征变化矢量,确定最优阈值,提取森林覆盖变化信息,并验证结果精度。【结果】(1)文中构建方法(PMKT-D)的试验结果总体精度为92.32%,Kappa系数为0.8437,结果精度显著优于对照方法;(2)2010−2020年间滇中东高原过渡山区森林覆盖变化面积356.88 km2,占影像总面积的1.06%,森林覆盖变化显著。【结论】基于多尺度分割,结合TCT和PCA能够有效消弱高原山区复杂地形和物候影响,发挥面向对象多特征变化矢量优势,增强森林覆盖变化光谱特征,显著提高变化检测精度。

[Objective]Based on multi-scale segmentation and tasseled cap transformation,this paper aims to develop a detection method for forest cover change in complex plateau area.[Method]Using Landsat 5 TM and Landsat 8 OLI multispectral images,taking the transitional mountainous area of the Yunnan Middle East Plateau of northwestern China as the research area,based on multi-scale segmentation,using the tasseled cap transformation(TCT)and principal component analysis(PCA),the object-oriented multi feature change vector was reconstructed,the optimal threshold was determined,forest cover change information was extracted,and the accuracy of the results was verified.[Result](1)The overall accuracy of the test results by PMKT-D was 92.32%,and the Kappa coefficient was 0.8437,which was significantly better than the control method.(2)From 2010 to 2020,the area of forest cover change in the transition mountain zone of Middle East Yunnan Plateau was 356.88 km2,accounting for 1.06%of the total image area,indicating significant changes of forest cover.[Conclusion]Based on multi-scale segmentation,the method combining TCT and PCA can effectively reduce the adverse effects of complex topography and phenology,enhance spectral characteristics of forest cover changes and improve the accuracy of the change detection significantly while take advantages of object-oriented multi-feature change vector.