The Research On Extracting The Changes Of Forest Landuse Based On ZY-1-02C Image And OLI Image

This research is mainly about detecting and extracting the changes of types of forest landuse. ZY-1-02 C data(earlier stage,2012) and Landsat 8 OLI data(later stage,2014) are used to detect and extract the changes of types of forest landuse in Longquan, Zhejiang province, and the accuracy of the results are verified.In this research, the images are processed by ArcGIS 10.1、ENVI 5.1、ERDAS IMAGE 9.2、eCognition 8.7、Matlab2012b、EXCEL2007 and other software. Firstly, we take some processes such as geometrical correction, image clipping, radial correction, image fusion. The accuracy of the classification based on pixels and Object-oriented method is compared. Then the changes are detected by different methods, such as pseudo color synthesis method, image subtraction and Object-oriented method. According to the detection results, extract the changes. Finally, we verify the accuracy of the results of extraction.The main content and results are as follows:(1) HSV is the best image fusion method in this research. A variety of methods are used for Image fusion. Then we use a number of indicators for the quantitative evaluation of the fusion results. The effect of HSV method is the best for both two images.(2) The accuracy of object oriented classification is higher than that of the pixel based classification method. The accuracy of random forest method is higher than that of other pixel based classification methods. In the classification accuracy evaluation of 02 C image, the accuracy of random forest method is 95.99%, the accuracy of maximum Likelihood method is 90.78%, the accuracy of parallel pipeline is 87.06%, the accuracy of mahalanobis distance method is 75.52%, the accuracy of support vector machine method is 92.51%. In the classification accuracy evaluation of OLI image, the accuracy of random forest method is 93.84%, the accuracy of maximum Likelihood method is 87.35%, the accuracy of parallel pipeline is 82.81%, the accuracy of mahalanobis distance method is 73.52%, the accuracy of support vector machine method is 92.13%. In object oriented classification results, the accuracy of 02 C classification is 98.55% and the accuracy of OLI classification is 98.08%.(3) The changes are detected by different methods, such as pseudo color synthesis method, image subtraction and Object-oriented method. Then the detected changes are extracted. Then changes error matrix is used for precision evaluation. According to the error matrix of pseudo color synthesis method, the missing rate is 78.32%, error rate is 9.83%, the overall accuracy is 78.80%; According to the error matrix of Image subtraction method, missing rate is 9.51% and the error rate is 51.23%, the overall accuracy is 72.50%; According to the error matrix of object-oriented method, the missing rate is 16.35% and the error rate is 15.06%, the overall accuracy is 91.80%.(4) From the precision evaluation of pseudo color synthesis method, we find that there are many omissions and less false changes. From the precision evaluation of image subtraction method, we find that there are less omissions and many false changes. From the precision evaluation of the object-oriented method, the omissions and false changes are the least of all three methods used in this research. So object oriented method is the best method in this research.(5) The results show that using data of ZY-1-02 C and Landsat 8 OLI data can extract the changes of forest landuse effectively. It proves that domestic satellites data can combine with foreign satellites data to monitor the changes of forest resources. The results broaden the applied range of multi satellite data.