فصلنامه سنجش از دور و GIS ایران
سال پنجم شماره 2 (پیاپی 18، تابستان 1392)

Today, aerial laser scanners (LiDAR) play an important role in 3D data acquisition from features. Filtering is one of the most affecting processes which are done on the LiDAR data. Filtering means distinguish between ground points and object points. Up to now, a lot of algorithms have been developed to automatic filtering of LiDAR data. Each of these algorithms has their own limitations and inefficiencies. One of the weaknesses of these algorithms is disability to remove the large buildings. This is due to their local-based operation. This paper presents a practical approach based on classification techniques to solve this problem. First, the LiDAR data are filtered using slope-based filtering algorithm which is one of the most known filtering algorithms. Afterward, the LiDAR range and intensity data are classified into five classes; road, tree, building, grassland and cement, using three classifiers; Support Vector Machine, Artificial Neural Network and Maximum Likelihood. Finally, the large building points, which had not been filtered by slope-based algorithm, are removed by used classification outputs. The results of accuracy assessment indicate that the Maximum Likelihood gives poor results, but Support Vector Machine and Artificial Neural Network present very good results. Generally, the usage classification techniques in order to improve the results of filtering algorithms causes a negligible increase in type I error and significant decrease in type II and total errors as well. Since, in filtering process, the type II error and total error are more important than type I error, performing of this supplementary process presents an effective result. Quantitative evaluation shows the output of the slope-based algorithm with 20º slope threshold after improving using Artificial Neural Network classifier, presents the best result. In this case type I error increased from 4.98% to 5.04%, type II error reduced from 9.043% to 4.49% and total error decreased from 7.03% to 4.76%.

Few researches have ever been done in allocation of demand points with a certain size services required to facility centers with a limited capacity services. Most proposed solutions for these issues are so complicated; therefore, some heuristic methods have been developed to deal with this subject based on the problem’s conditions. Although in many allocation problems a major part of information and analysis is spatial, operational research and science management researchers have paid a little attention to this fact. Allocation of injuries to medical centers is one of vital tasks in earthquake disaster management. Developing and using an optimized method to allocate injuries to medical centers, definitely decreases the rescue time and could highly mitigate the number of human losses. In this paper, we created a heuristic method based on genetic algorithm and geospatial information system to optimize earthquake injurie's allocation to medical centers. To assess the qualities of proposed algorithm, two different scenarios about the numbers and spatial distribution of earthquake injuries have been analyzed in the study area. The effect of initial population size parameter of genetic algorithm in solving the problem was evaluated as well. The results of this study showed that developed framework with a proper initial population size could be an appropriate solution for optimizing these types of allocation problems.

The aim of this study is statistical modeling of rainfed wheat yield in Kurdistan province of Iran. In order to do this, multiple linear regression models were used. Dependent variable was rainfed wheat yield and independent variables were three, including climatic, agroclimatic and spectral indices extracted from satellite images. Boot strop method was run on all calculated models and based on this method forecasting values were presented for 2003 to 2006. The results show that the best model is based on Reproductive stage data. In Divandareh County, the best model is based on second growing period and creative process. In Marivan County, the best model established on dormancy stage and in Sanandaj and Ghorveh County calculated based on all growing season data. Also results show that among spectral indices, DVI index is a better predictive variable. Furthermore the accuracy of calculated models is increases based on a combination of agroclimatic and remote sensing indices. Running bootstrap resampling method also increases accuracy of calculated models. Therefore this method can be used for rainfed wheat yield forecasting.

This research is to investigate the possibility of porphyry copper mineralization in Hanza Mountain near Darb-e-Behesht located in southern section of Dehaj-Sardooiyeh zone. Porphyry copper mineralization is known to occur in Rameshk, Bandar-e-Hanza, Soorakh-e-Maar and Garu. Spectral angle mapped (SAM) and logical operators were used to identify phyllic and argillic alteration halos that usually occur around porphyry copper deposits. Logical operators are in fact a set of band ratios, such that in first step vegetation canopy is eliminated and then alteration halos are extracted and distinguished from each other. In SAM, based on spectral signatures of altered minerals in SWIR wavelength, argillic and phyllic alterations were identified the presence of kaolinite and muscovite. Similarity coefficient of both methods is medium and similarity coefficient of mineralized points that is verified by both methods is high. Therefore, the areas where both SAM and logical operators verify each other have the highest possibility of porphyry copper mineralization.

Forest fire is a phenomenon that causes many economic and lifelong losses. Identification and assessment of parameters that effect on fire occurrence can be useful in fire prevention and reduction of related losses. In this research, high fire potential areas of Golestan were identified using artificial neural network (ANN) and multi-criteria decision making analysis. In the proposed method, at first the static fire risk was obtained using multi-criteria decision making analysis and then ANN usage for dynamic fire risk and finally the fire risk was obtained through linear combination of these two maps. Each pixel in final risk map shows the probability of fire occurrence. The June 2005 data was used to test the proposed model. For this purpose the proportion of occurred fire percent in the region, was identified as dangerous to its corresponding area size described as accuracy criteria. This criterion was obtained 1.86 for parameters with equal weight and obtained 2.21 when the weights derived through AHP analysis which indicates accuracy improvement. According to results, 70.49 percent of fires occurred in the areas that was identified as dangerous region while no fire occurred in non-danger and low danger region, and 29.51 percent of fires occurred in the areas, was identified as medium-danger region. Also the accuracy criterion was obtained 2.77 for final model.

Different algorithms have been presented for Pan Sharpening in remote sensing high-resolution images. In order to select the appropriate Pan-sharpening algorithm, a criterion is needed for comparing Pan-sharpening algorithms that would investigate Pan-sharpening more accurate algorithm compared with other Pan-sharpening algorithms. Several statistical indexes are presented for Pan-sharpened images of spectral and spatial quality assessment. One of the limitations of these spectral quality assessment indexes is that, they measure the average quality based on single pixels and do not consider the neighborhood of the pixel and the image details, while the limitation of the spatial quality assessment is that the point detection filter is located at 3*3 neighborhoods. In this study, the wavelet transform and features (point, line and edge) detection filters are categorized as indexes for spectral and spatial quality assessment of Pan-sharpened images in order to justify this title. Both the proposed indexes for quality assessment and its neighborhood are applied on each pixel. The spatial and spectral quality assessment is designed to perform on 5*5 neighborhoods of any pixels. The proposed indexes are evaluated quantitatively in 13 test sites of different urban land covers.

During recent years, dust storms are becoming very dangerous crisis with regional dimensions. Khuzestan province is one of these areas which is affected by this phenomenon. Meteorological and vegetation parameters play important roles in causing dust storms. In this paper, the relationship between spatial-temporal changes of these parameters and dust storm frequency (DSF) has been modeled. For this purpose, climatic data (wind speed, temperature, precipitation and humidity) and DSF were evaluated since 2000 to 2008. During these years, NDVI was extracted monthly by using MODIS images and then interpolating of climatic data was done by Kriging method. After preprocessing, data were pixeled to 18*19 matrix elements. A multivariate linear regression model (in terms of parameter) was estimated between climatic and vegetation parameters (independent variables) and DSF for each pixel (matrix element). Base on the models, except for month June, the accordance between predicted temporal values and actual values of dust events were too strong (Corr=96%). Soil, land use and vegetation land cover maps showed the high rate of potential of soil dispersion in south of Khuzestan which produced the most sources of domestic dust storms in this region. The accuracy of model was evaluated for 10 classes of soil, separately. The highest and lowest accuracies were for salty and marshes and sandy land, respectively. So, destruction of vegetation cover effectively cause increasing domestic source dust storms in Khuzestan and saline and alkaline soils intensify the emission of dust in the atmosphere.