基于TM影像的绍兴地区土地利用变化监测

Monitoring Land-Use Change in Shaoxing based on TM Images

以1984年和1997年两期TM影像,结合相关辅助信息,进行绍兴试验区的土地利用变化监测研究。在两时相影像校正处理基础上,进行影像的光谱数据转换处理并通过两时相影像的亮度轴(B)、绿度轴(G)和湿度轴(W)3个分量的差值运算进行信息复合。选取影像中的典型变化区并调整变化范围和步长搜寻获取变化和非变化像元的最佳变化强度阈值,采用直接比较分析法在掩膜的ΔBΔGΔW空间中提取土地利用变化信息,从而避免了分类后比较法的误差累积并出现不合理变化类型的缺陷。在此基础上利用地物时相特征和地理辅助数据,通过综合应用地物的光谱、空间和时相信息并以知识推理的方式对前期单一分类方法的结果进行修正处理。精度分析表明采用本方法能较为可靠地进行土地利用变化监测。

In this study, multi-temporal TM images （in 1984 and 1997） with geographical ancillary data were used to monitor land-use change in Shaoxing. Based on the images pre-processing procedures, especially in geometric rectification and relative radiometric normalization, one image was employed as the reference scene to which other scenes were registered. Land-use change information was acquired mainly by using two methods. One is post-classification image comparison （map-to-map comparison） which requires that images from two dates are classified individually, and then the results are acquired by comparing two classified maps. The other is direct classification of stacked multi-temporal images （image-to-image comparison） which requires comparison images from two dates, and land-use change information is directly acquired by classifying the multitemporal images. The focus of the study is to extract land-use change information by means of image-to-image comparison technique. Data pre-processing can improve the classification accuracy of land-use and land-use change information. Because of unavailable parameters （such as sensor calibration parameters, atmospheric corrections parameters and solar elevation angle parameters, et al. ） and applicability of the correction model for the absolute radiometric calibration, the absolute radiometric calibration is difficult to be used. The Relative radiometric normalization correction is an effective correction method to be used usually. The relative radiometric normalization correction is necessary when the direct classification of stacked multitemporal images is used to extract land-use change information. The Tasseled Cap transformation is used to reduce data, depress noise and enhance information. Through the Tasseled Cap transformation, the bands of TM1-5 and TM7 are transformed to three planes： Brightness （B）, Greenness （G） and Wetness （W）. Different land-use types occupy distinct spectral locations in BGW space. As land-use changes over time, a change in reflectivity of the surface exhibits corresponding changes in BGW values. The data from different periods were compounded to three-dimensional △B△G△W space with minus operation. In △B△G△W space, different land-use change types occupy distinct spectral locations. The classification accuracy of image can be improved by determining change magnitude threshold reasonably from the compounding image. Typical land-use change areas of the image were selected to determine change magnitude threshold reasonably by rectifying flexible range and pace from the compounding image. The land-use change types were directly acquired by maximum likelihood classification of the masked △B△G△W image. Because of complexity of land-use/cover and spectral confusion, It＇s difficult to accurately classify remote sensing image relied solely on the spectral information only using one classification method. Knowledge-based image classification procedure based on temporal, spatial, spectral characteristics is an effective approach to accurately reclassify the image through knowledge reasoning in post-classification processing. In post-classification processing, knowledge reasoning technique is used to accurately reclassify the map with temporal characteristics of land-use types and geographic ancillary data, and the errors can be substantially reduced with the knowledge-based reclassification procedures. The results proved this technique applicable to detect land-use change from TM image.