SPECTRAL-SPATIAL CLASSIFICATION OF HYPERSPECTRAL IMAGERY USING NEURAL NETWORK ALGORITHM AND HIERARCHICAL SEGMENTATION
D. Akbari a , M. Moradizadeh b, M. Akbari c
a Department of Surveying and Geomatics Engineering, College of Engineering, University of Zabol, Zabol, Irandavoodakbari@uoz.ac.ir.
b Department of Geomatics, Faculty of Civil and Transportation Engineering, University of Isfahan, Isfahan, Iran
c Department of Civil Engineering, College of Engineering, University of Birjand, Birjand, Iran
Commission II, WG II/5
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-2/W12, 2019
Int. Worksh. on “Photogrammetric & Computer Vision Techniques for Video Surveillance, Biometrics and Biomedicine”, 13–15 May 2019, Moscow, Russia
Studies on the change in occupation and land-use are of great importance in order to understand landscape dynamics in the process of agricultural land degradation, urbanization, desertification, deforestation and all change in the landscape global of a region. The most effective procedure to measure the degree of land-cover and land-use changes is the multi-date study. For this purpose, the aim of this work is to analyze the current evolution of land-use and land-cover (LULC) using remote sensing techniques, in order to better understand this evolution. For this purpose, a diachronic approach is applied to satellite images acquired in 1987 to 2018 of Ma'rib city Yemen. The LULC maps we obtained were produced from different image analysis procedures (non-supervised classification and recode technique) to map the land-use and land-cover. The objective of this study is to apply reproducibly and generalizable a predefined nomenclature to different scenes of satellite images. The first step consists in interpreting the radiometric classes obtained by non-supervised classification so as to form the classes of the thematic nomenclature. An improvement of the classification is then obtained by using the recode technique which makes it possible to correctly reassign the previously badly classified pixels of the satellite images classification. Land-cover maps obtained from remote sensing were used to quantify the rate of change (Tc) and (Tg) of area occupied by each class. The results will indicate the most changeable period and the percentage of overall change in the study area (Ma'rib Yemen), and helped to identify and characterize the spatial and temporal evolution of land use in the district over a period of thirty-one years (1987 to 2018). They reveal that annual average rates of decline for the water body is -83.5% and -9.96% for the sandy land. However, it was observed an increase in built-up area 365.52% and farm land 324.52% classes.
KEY WORDS: Remote sensing, Hyperspectral image, neural network, Hierarchical segmentation, Marker selection