隨著衛星影像的持續接收，利用衛星影像進行土地變遷偵測更趨頻繁，為使變遷偵測朝向高精度及高效率，變遷偵測的方法不斷的提出。本論文使用多時期分類法進行變遷偵測，以線性複合模式作為分類器，最小二乘子空間投影法作為求解方式，產生變遷類別影像，稱為單層次線性複合模式變遷偵測法。但由於單層次線性複合模式具有變遷組合類別數必須小於合併影像波段數限制，因此本論文進一步以多層次(Multi-Level)線性複合模式進行變遷偵測。本論文測試3組影像，使用多層次線性複合模式進行變遷偵測，其模擬影像變遷偵測整體精度達到90%以上，SPOT衛星影像變遷偵測整體精度達到80%以上。因此預期多時期衛星影像，以複性複合模式作為變遷偵測方式，不失為一個可實際應用的方法。 The usage of satellite images for land cover change detection has been an important task for environment monitoring. In this paper, we use multi-temporal satellite images and classifier to detect change regions. The classifier is Linear Mixing Model (LMM) with Least Square Orthogonal Subspace Projection (LSOSP). LMM is a model to descript classes in the image, and LSOSP is one of the methods to solution the LMM. It is proposed to detect the signal of the desired land-cover materials and eliminate the undesired signatures. Finally, an intensity image would be obtained to represent the intension of the desired signatures. However, this method cannot discriminate classes more than the number of bands of the combined image. Therefore, we proposed multi-level linear mixing model to solve this problem. The test data of this study include one simulation image and two SPOT4 satellite images. The overall accuracy is about 80%, and the kappa coefficient is about 0.6. Simulated data and real SPOT images are used for testing, and the results indicate that change detection using LMM is workable.
