Object-oriented land cover classification using multi-temporal HJ-1 CCD imagery: A case study in central Shandong province, China

This paper focuses on object-oriented land cover classification using multi-temporal remotely sensed imagery. We proposed an approach by building rules using multi-temporal HJ-1 CCD imagery and other auxiliary data to classify various land cover types in central Shandong province. We analyzed the seasonal dynamics of vegetation indices (EVI (Enhanced Vegetation index) and NDVI). Vegetation index time series of multi-temporal images can help differentiate forest types. Given the difficulties of vegetation classification, especially in mountainous area, more information available such as DEM, slope, spatial features and priori knowledge were also utilized. The overall accuracy and Kappa coefficient of land cover classification are 80.1% and 0.76, respectively. The results show that besides the spectral information, texture, DEM, slope and auxiliary data are very useful for land cover classification. Multi-temporal information can improve the vegetation classification result significantly and meanwhile has much potential to be explored.