Detecting the changes in rural communities in Taiwan by applying multiphase segmentation on FORMOSA-2 satellite imagery

12 Agricultural activities mainly occur in rural areas; recently, ecological conservation 13 and biological diversity are being emphasized in rural communities to promote 14 sustainable development for rural communities, especially for rural communities in 15 Taiwan. Therefore, since 2005, many rural communities in Taiwan have compiled 16 their own development strategies in order to create their own unique characteristics to 17 attract people to visit and stay in rural communities. By implementing these strategies, 18 young people can stay in their own rural communities and the rural communities are 19 rejuvenated. However, some rural communities introduce artificial construction into 20 the community such that the ecological and biological environments are significantly 21 degraded. The strategies need to be efficiently monitored because up to 67 rural 22 communities have proposed rejuvenation projects. In 2015, up to 440 rural 23 communities were estimated to be involved in rural community rejuvenations. How to 24 monitor the changes occurring in those rural communities participating in rural 25 community rejuvenation such that ecological conservation and ecological diversity 26 can be satisfied is an important issue in rural community management. Remote 27 sensing provides an efficient and rapid method to achieve this issue. Segmentation 28 plays a fundamental role in human perception. In this respect, segmentation can be 29 used as the process of transforming the collection of pixels of an image into a group 30 of regions or objects with meaning. This paper proposed an algorithm based on the 31 multiphase approach to segment the normalized difference vegetation index, NDVI, 32 of the rural communities into several sub-regions, and to have the NDVI distribution 33 in each sub-region be homogeneous. Those regions whose values of NDVI are close 34 will be merged into the same class. In doing so, a complex NDVI map can be 35 simplified into two groups: the high and low values of NDVI. The class with low 36 NDVI values corresponds to those regions containing roads, buildings, and other 37 manmade construction works and the class with high values of NDVI indicates that 38 those regions contain vegetation in good health. In order to verify the processed 39 results, the regional boundaries were extracted and laid down on the given images to 40 check whether the extracted boundaries were laid down on buildings, roads, or other 41 artificial constructions. In addition to the proposed approach, another approach called 42 statistical region merging was employed by grouping sets of pixels with homogeneous 43 properties such that those sets are iteratively grown by combining smaller regions or 44 pixels. In doing so, the segmented NDVI map can be generated. By comparing the 45 areas of the merged classes in different years, the changes occurring in the rural 46 communities of Taiwan can be detected. The satellite imagery of FORMOSA-2 with 47 2-meter ground resolution is employed to evaluate the performance of the proposed 48 approach. The satellite imagery of two rural communities (Jhumen and Taomi 49 communities) is chosen to evaluate environmental changes between 2005 and 2010. 50 The change maps of 2005 and 2010 show that a high density of green on a patch of 51 land is increased by 19.62 hectares in Jhumen community and conversely a similar 52 patch of land is significantly decreased by 236.59 hectares in Taomi community. 53 Furthermore, the change maps created by another image segmentation method called 54 statistical region merging generate similar processed results to multiphase 55 segmentation. 56