Behnam Asghari Beirami; Mehdi Mokhtarzade
Abstract
The use of spatial features to improve the classification accuracy of hyperspectral images has become popular in recent years. Various methods for spectral-spatial classification of hyperspectral images have been introduced to date, and relevant research is being conducted to develop methods with a more ...
Read More
The use of spatial features to improve the classification accuracy of hyperspectral images has become popular in recent years. Various methods for spectral-spatial classification of hyperspectral images have been introduced to date, and relevant research is being conducted to develop methods with a more straightforward structure and higher accuracy. This paper introduces a new method for producing efficient features for classifying hyperspectral images based on combining extracted features from the weighted local kernel matrix of spectral and fractal features. One of the main advantages of weighted local kernel matrices is that they model nonlinear dependencies between features that are not taken into account by traditional feature generation methods. In this study's proposed method, the weighted kernel local matrix method is used in order to generate new features from spectral features and directional fractal dimension features. Then these two feature vectors are joined together for each pixel and form a vector rich in spectral-spatial information. Finally, to determine each pixel's label, the obtained feature vector is classified by the support vector machine (SVM) algorithm. The results obtained from two real hyperspectral images of Indian Pine and Pavia University show that the accuracies in the proposed method are above 98% on average in both data sets, which is more than 5% higher than the average accuracy of several other hyperspectral image classification methods.
Maryam Teimouri; Mehdi Mokhtarzade; Mohammad Javad Valadan Zoej
Abstract
In this study, the SAR data is used as a supplementary data to overcome the limitations of the multispectral (MS) image in building detection. Therefore, the proposed method utilizes a multisensor data fusion to take the advantages of both MS and SAR data together. In addition, two different filter-based ...
Read More
In this study, the SAR data is used as a supplementary data to overcome the limitations of the multispectral (MS) image in building detection. Therefore, the proposed method utilizes a multisensor data fusion to take the advantages of both MS and SAR data together. In addition, two different filter-based feature selection methods, MNF and PCA, are investigated as an alternative scenario when the training data is not accessible. In this respect, the optimum feature vector is selected using MNF, PCA and Genetic methods from MS and SAR data, separately. Thereafter, each selected feature vector is used to classify the images by implementing the support vector machine (SVM) and the artificial neural network classification methods. The experimental result shows that the PCA is able to select the feature vector without the need of training data as well as genetic algorithm. However, the MS classification result is poor where both roofs and streets are covered with asphalt. In this framework, the fusion of SAR and MS images in feature level was utilized to improve the classification results. Finally, to assign a label at the sample, a majority voting is calculated between the used classification methods results. However, according to the noisy result, using the neighborhood information in the form of a moving spatial window in different sizes is examined to determine the label of the central pixel more accurately. According to the experimental results, the overall accuracy and building detection accuracy are obtained 92.82% and 80.14%, respectively, which represent the satisfying performance of the proposed method.
Melika Haghparast; Mehdi Mokhtarzade
Volume 10, Issue 1 , June 2018, , Pages 91-108
Abstract
Due to the global scope of water resources, ground measurements of the quality parameters are not feasible, as well as traditional sampling of water and laboratory analysis is very costly and time-consuming. In studies, estimation of turbidity and chlorophyll a concentrations as the most important water ...
Read More
Due to the global scope of water resources, ground measurements of the quality parameters are not feasible, as well as traditional sampling of water and laboratory analysis is very costly and time-consuming. In studies, estimation of turbidity and chlorophyll a concentrations as the most important water quality parameters using artificial neural networks have been done by researchers. Considering the difficulties in providing a high number of training data in aquatic environments, the use of more robust hybrid networks such as the wavelet neural network is suggested. In this research, various types of wavelet functions were used as a network activation function, and the best network was used to estimate chlorophyll a and turbidity respectively, wavelet neural networks with a Morelt and a Mexican hat activation function, the data used for the reflection of the ocean reflectance of the modis sensor, Due to the use of multi-time images, the radiometric normalization of data was done and the results were significantly improved compared to the time when the non-normalized images were used. in addition to increasing the number of training data, the network generalization capability is provided to other days, and the accuracy of the network in this case increased compared to the one-day condition. the RMSE for the best model to estimate turbidity And chlorophyll a concentration was 0.047 and 0.071, respectively, which is acceptable in comparison with field accuracy of 0.1, and can be a alternative method for field measurements.
F Mohseni; M Mokhtarzadeh
Volume 9, Issue 4 , May 2017, , Pages 1-22
Abstract
Soil moisture plays an important role in interactive processes between earth and atmosphere and global climate changes. In recent decades, there has been a great research interest to determine soil moisture from remote sensing methods. Triangular or trapezoidal methods are the most common remote sensing ...
Read More
Soil moisture plays an important role in interactive processes between earth and atmosphere and global climate changes. In recent decades, there has been a great research interest to determine soil moisture from remote sensing methods. Triangular or trapezoidal methods are the most common remote sensing methods that apply the combination of thermal and optical satellite images to estimate soil moisture content. The accuracy of methods governed by the accuracy of saturated and dry edges that define from vegetation/ temperature scatter plot. A main limitation of these methods arose in some days or in some vegetation condition that dry and wet edges cannot be defined correctly. This concern is addressed in this paper by using the temperature and vegetation information during one year interval to form the temperature-vegetation scatter plot, saturated edge and dry edge exactly. The main contribution of the paper is, however, the introduction of co-moisture lines in the one-year scatter plot. These lines are later applied to define the wet and dry edges of each individual day which are taken as the two closest co-moisture lines that contain all corresponding pixels of that day. The soil moisture index as a parameter dependent to evaporation efficiency is finally estimated from the slope and intercept of these two co-moisture lines. The proposed soil moisture index calculated from co-moisture was implemented and validated in Manitoba, Canada area while MODIS satellite images, taken in 28 cloudless days of year 2014, were used as the input data. The correlation between ground soil moisture data and proposed soil moisture index was estimated. Correlation of 0.92 was achieved for low vegetation days and lower in days with higher vegetation densities.
Volume 7, Issue 1 , December 2015, , Pages 81-94
Abstract
Classification is one of the most widely used remote sensing analysis techniques. In the conventional remote sensing supervised classification, training information and classification result are represented in a one-pixel-one-class method. Fuzzy methods have been widely applied in image classification, ...
Read More
Classification is one of the most widely used remote sensing analysis techniques. In the conventional remote sensing supervised classification, training information and classification result are represented in a one-pixel-one-class method. Fuzzy methods have been widely applied in image classification, which are believed to be more appropriate for handling uncertainty and mixed pixels in remote sensing. Also recent researches show that using neighborhood information with spectral information lead to higher accuracy in classification. Due to the dependence on initial classifier,the use of neighborhood information in the post processing of classification results is one of the reasons for its use in this research. Connectivity rules in fuzzy topological space are one of methods for using neighborhood information in post processing step. In case of using more than one classifier, it is possible to integrate the results. In this research two methods have been proposed for spatial integration results by using connectivity rules in fuzzy topological space. In first method, one of the two classifiers will be based and in second method, only pixels that are classified in the same manner in both and simultaneously not boundary pixel, will keep their own labels in final image. The results show that first method Provides better accuracy compared with second method and generally accuracy is improved when spatial integration results is used in compare with using only one classifier. The maximum overall accuracy and overall kappa values are obtained respectively 89.01 and 88.98 when maximum likelihood classifier is based in first method. Keywords: Fuzzy Classification, Fuzzy Topological Space, Integration, Connectivity Rules.
A Baghani; M.J Valadan Zoej; M Mokhtarzade
Volume 7, Issue 2 , November 2015, , Pages 1-18
Abstract
Due to the absence of either satellite ephemeris information or camera model for various high resolution satellite images, rational functions models (RFMs) are widely used by photogrammetric and remote sensing communities. This method has various disadvantages such as: The dependency of this method on ...
Read More
Due to the absence of either satellite ephemeris information or camera model for various high resolution satellite images, rational functions models (RFMs) are widely used by photogrammetric and remote sensing communities. This method has various disadvantages such as: The dependency of this method on many ground control points (GCPs), numerical complexity and particularly terms selection. As there is no physical meaning for the terms of RFM, in traditional solution all of them are involved in the computational process which causes over-parameterization. In this letter, a modified Ant Colony Optimization is applied to identify the optimal terms for RFMs. For this purpose this method is tested on three images with different geometric correction levels, different coordinate systems (UTM, CT & Geodetic) and different combination of Ground Control Points (GCPs) and Independent Check Points (ICPs), without normalization of the image and ground coordinates. Experimental results demonstrate how well the proposed algorithm can determine an RFM, which is optimal in both the total number of terms and the positional accuracy. The results have showed that the CT coordinate system has the better capability in accuracy and convergence’s speed. As a conclusion, ACO when using for RFM optimization, can achieve subpixel accuracy even with just four GCPs.
Jadidi Milad Niroumand; Mehdi Mokhtarzade; Mahmood Reza Sahebi
Volume 7, Issue 3 , November 2015, , Pages 1-16
Abstract
The mixed pixels are considered as a major challenge in land cover mapping procedure from satellite imagery. Developments of the spectral unmixing and soft classification methods have provided the possibility for estimation of class proportions within the pixels. However, sub-pixel land cover mapping ...
Read More
The mixed pixels are considered as a major challenge in land cover mapping procedure from satellite imagery. Developments of the spectral unmixing and soft classification methods have provided the possibility for estimation of class proportions within the pixels. However, sub-pixel land cover mapping requires the spatial allocation of the sub-pixels. Recently, the Super Resolution Mapping (SRM) techniques have been developed for optimization of the sub-pixels spatial arrangement using the outputs of soft classifiers and based on the concepts of spatial dependency. In this research, the overall capability of the simulated annealing algorithm was evaluated through sub-pixel land cover mapping of the study area. To do so, a novel method was proposed for generating new solutions in each step of the algorithm and then the results were compared to the traditional method. On the other hand, the effective parameters on the performance of the algorithm (e.g. zoom factor, cooling function type, static and dynamic iterations) were investigated. According to the obtained results, higher values of zoom factor yields more promising overall accuracy . Also, the geometric function was found as the optimal cooling function with respect to the overall accuracy and processing speed. Meanwhile, dynamic iterations demonstrated more accuracy than the static case. As another key result of the paper, the proposed method for generating the new solutions in simulated annealing algorithm is led to increasing of the overall accuracy and also reducing the processing time of algorithm up to 50 percent. The most accurate result of the proposed algorithm, which was obtained for the that case of being independent from soft classifier, is determined 94.97 percent
Fateme Ameri1; Mohammad Javad Valadan Zoej; Mehdi Mokhtarzade
Volume 7, Issue 3 , November 2015, , Pages 33-48
Abstract
Nowadays, extraction of roads from digital aerial and satellite images is a common method of road database construction. Regarding to massive amount of road data and time and cost effective updating requirements, automation procedure is becoming an essential. In this research, which is mostly concentrated ...
Read More
Nowadays, extraction of roads from digital aerial and satellite images is a common method of road database construction. Regarding to massive amount of road data and time and cost effective updating requirements, automation procedure is becoming an essential. In this research, which is mostly concentrated on road vectorization process, an automatic approach of road centerline vectorization from detected road image with negligible operator interventions is designed. The proposed system consists of two main stages including road key points determination and connection. At the first stage, the road key points representative of the road centerline are determined using particle swarm optimization clustering. At the second stage, in order to model the road networks weighted graph theory is considered. In this model cost of each connection is calculated by aggregating appropriate road geometric criteria by means of ordered weighted averaging operators. The least cost connections constitute the vectorized road networks. The proposed approach was implemented on several high resolution satellite images and their results were compared with the results of the minimum spanning tree algorithm. On the whole, the obtaining results proved the efficiency of the vectorization approach in attaining the complete and accurate road network. Extracting different road shapes including direct and curved roads, roads with different widths, parallel roads with different distances, junctions and square with average RMSE value about 0.9 meter, average completeness of %94, and average correctness greater than %95 proves the efficiency of the algorithm in yielding complete road networks.
Volume 6, Issue 4 , October 2014
Abstract
Land cover information is one of the most important prerequisite in urban management system. In this way remote sensing, as the most economic technology, is mainly used to produce land cover maps. Considering the complicated and dense urban areas in third world countries, object based approaches are ...
Read More
Land cover information is one of the most important prerequisite in urban management system. In this way remote sensing, as the most economic technology, is mainly used to produce land cover maps. Considering the complicated and dense urban areas in third world countries, object based approaches are suggested as an effective image processing technique. The purpose of this paper are the introduction of a new object based approach for classification of complicated urban area using high resolution satellite image and approaching to a standard and effective process of map generation by satellite images. This paper used a new approach to select the segmentation parameters and a new hierarchical classification model based on a rule based strategy is used to overcome the confusions between urban classes too. In this article an innovative hierarchical model is proposed for object-based classification of complicated urban areas. In this way, beside of feature space optimization in a multi scale analysis, rule based and fuzzy nearest neighbor approaches are used as the object-based classification strategies. The proposed method is implemented on an urban IKONOS image where 84% and 87%overall accuracies are obtained for rule based and fuzzy nearest neighbor classification approaches respectively. The implementation of the devised algorithm on another IKONOS image moved its general ability to other urban areas. Keywords: Land cover classification, Rule based, Object based, Fuzzy nearest neighbor, Complicated urban areas.
Volume 6, Issue 1 , April 2014
Abstract
Nowadays, urban land use and land cover information at the micro and macro levels of planning have a particular importance. So many researches have not been done in land use information extraction. Remote sensing as an inexpensive and fast method, and particular with appearance of object-based analysis, ...
Read More
Nowadays, urban land use and land cover information at the micro and macro levels of planning have a particular importance. So many researches have not been done in land use information extraction. Remote sensing as an inexpensive and fast method, and particular with appearance of object-based analysis, has an appropriate potential for this. In this paper, the aim is land use information extraction on a dense and complicated urban area. For this purpose, a hierarchical system inclusive land cover and land use levels has been used. After the implementation of a step by step land cover classification approach, land use unites extraction are done. In the next stage, feature space inclusive more than 50 conceptual features based on land cover information is designed and extracted. After this stage, optimized features among these features with high separability using SFFS are extracted. Finally a fuzzy nearest neighbor classification for land use classification based on optimized is implemented. Land use classification is performed on two combined and uncombined class system that combined class is recognized as most appropriate class system. In the present approach without considering area criteria of land use object, 82% overall accuracy and with this criteria 85% overall accuracy is achieved.
Volume 4, Issue 3 , September 2012
Abstract
This paper proposes a new approach for geometrical modeling of satellite imagery which uses 2D-polynomials for 3D point determination from satellite stereo images. In this model, 2D polynomials are considered as additional parameters in colinearity equation, instead of considering as models for relating ...
Read More
This paper proposes a new approach for geometrical modeling of satellite imagery which uses 2D-polynomials for 3D point determination from satellite stereo images. In this model, 2D polynomials are considered as additional parameters in colinearity equation, instead of considering as models for relating between the ground and space images. Orbital parameters model are used as fundamental colinearity equations in this modeling. Essential parameters in the orbital parameters model are determined from satellite ephemeris data and they are considered as fixed parameters in the modeling. In this model, polynomial coefficients are the only unknown parameters which are determined from GCPs in a linear equations set. The major advantages of this model are: Decreasing the performance complexities in using orbital parameters model, ease of implementation, applicability on raw geometrically corrected images, and possibility of using maximum capacity of satellite auxiliary data, and linearity of equations in space intersection procedure. Implementation of this model on different datasets shows high potentiality of the mentioned approach for 3D point determinations from satellite stereo images.