Accurate estimate of the land cover is of particular importance in environmental management and planning. Different methods have so far been developed for the classification of satellite imagery. Regarding the weakness of the hard classifiers in the presence of the mixed pixels, soft classification methods have been designed to estimate the fractions of different classes within mixed pixels. Linear spectral mixture model could be considered as the most popular method of soft classification. The accuracy of this model is extremely depended on the quality of the input pure pixels. The extraction of these pixels is often hard and challengeable. In this paper, making use of physical parameters (brightness, wetness, and greenness) is proposed for extraction of pure pixels. The evaluation criteria of unconstrained linear spectral mixture model were used assess the quality of pure pixels extracted from the proposed method. By Implementing of this model on the satellite image of the study area, the total number of under-shoot and over-shoot pixels was determined as 4.43 percent of the total image pixels. Also RMSE equaled to 2.68. Consequently, the proposed method demonstrated promising accuracy in the results of the linear spectral mixture model without considering any constraint.    

Keywords: Land cover, Spectral mixture model, Pure pixel, Physical parameter

Full-Text [PDF 584 kb]   (4249 Downloads)