TY - JOUR JF - kntu-jgit JO - jgit VL - 6 IS - 4 PY - 2019 Y1 - 2019/3/01 TI - Land Cover Subpixel Change Detection using Hyperspectral Images Based on Spectral Unmixing and Post-processing TT - آشکارسازی تغییرات زیر پیکسلی کاربری اراضی در تصاویر ابرطیفی مبتنی بر جداسازی طیفی و پس‌پردازش N2 - The earth is continually being influenced by some actions such as flood, tornado and human artificial activities. This process causes the changes in land cover type. Thus, for optimal management of the use of resources, it is necessary to be aware of these changes. Today’s remote sensing plays key role in geology and environmental monitoring by its high resolution, wide covering and low cost of data receiving from the earth and it has many applications such as change detection. To manage the resources optimally, in local and gloal scale, accuracy and being on-time are very substantial. Hyperspectral images, with thier high ability of spectral resolution, can improve change detection in result and extract more detail of changes. In this research a new method of change detection for hyperspectral imagery using the Image-Differencing, Otsu and spectral unmixing algorithms is presented . The proposed method is presented in three steps: (1) Data correction using image differencing method and data conversion to new computing space. At this space, the changed areas would be more outstanding compare to previous space. (2) the decision about the nature of endmembers is made using Otsu algorithm. (3) spatial resolution enhancement based on abundance map. The proposed method can automatically extract binary change map. In addition, this method provides information about the nature of change in sub-pixel level. To examine the performance of the proposed method, the hyperspectral imagery (by Hyperion sensors) from Chiangsu fields in china and a simulated data from the AVIRIS sensor were used. The results show the high accuracy of the proposed method in comparison with other methods. Its overall accuracy is more than 93% and its kappa coefficiency is 0.85 and mean false alarm rates is under 7% for China dataset. And also, the results for second dataset are as follow: the overall accuracy is more than 99% and kappa coefficiency is 0.82 and mean false alarm rates is under 1%. SP - 97 EP - 117 AU - Hasanlou, Mahdi AU - Seydi, Seyed Teimoor AD - University of Tehran KW - Unsupervised Change Detection KW - Spectral Unmixing KW - Hyperspectral Images KW - Otsu Algorithm KW - Sub-pixel. UR - http://jgit.kntu.ac.ir/article-1-646-en.html DO - 10.29252/jgit.6.4.97 ER -