جستجوی مقالات مرتبط با کلیدواژه « Hyperspectral Image » در نشریات گروه « عمران »

Information about vegetation cover and their health has always been interesting to ecologists due to its importance in terms of habitat, energy production and other important characteristics of plants on the earth planet. Nowadays, developments in remote sensing technologies caused more remotely sensed data accessible to researchers. The combination of these data improves the object classification and recognition results. Recently, hyperspectral and Lidar data has been used for vegetation covers classification. The spectral information derived from hyperspectral data is used to classify and identify the vegetation cover. However; due to the spectral similarities between various vegetation types, false positive results are increased. Using relief information extracted from Lidar data can solve these kinds of errors and can be very efficient for improving the object recognition results. Spectral similarities and spatial adjacencies between various kinds of objects, shadow and occluded areas behind high rise objects as well as the complex relationships between various object types lead to the difficulties and ambiguities in vegetation recognition among other objects in urban areas. Therefore, new procedures and higher levels of modifications should be considered for improving the object recognition results. In recent years, the multi-agent systems have been considered as one of the most powerful tools for solving the problems of automatic object recognition in urban areas.

According to the difficulties of vegetation recognition in complex urban areas, the proposed object recognition in this paper is a decision level fusion strategy between hyperspectral and Lidar data based on the capabilities of the multi-agent systems. Vegetation indices from hyperspectral image are used as spectral features in the knowledge base. Moreover, digital surface model which is produced from Lidar data is used for height features extraction. After producing a rich knowledge base containing the spectral and height based features, the proposed hierarchical classification is performed which is composed of two steps; step 1: initial vegetation candidate recognition, step 2: vegetation classification based on the capabilities of the multi-agent systems. Applying the optimum thresholds on the normalized difference vegetation index in the first step produces a binary image containing the initial vegetation candidates. The multi-agent system in the second step of the proposed method in this paper contains several object recognition agents (one agent per each vegetation cover type), a coordinator agent and a yellow page. The object recognition agents have three layered internal architecture and use the belief-desire-intention (BDI) reasoning model. 

The capabilities of the proposed multi-agent vegetation recognition algorithm in this paper is evaluated based on the hyperspectral and Lidar data collected from the University of Houston and the surrounding areas. Four object recognition agents are defined for trees, healthy grass, water-stress grass and artificial grass. These four object recognition agents perform their reasoning based on the pre-defined spectral and height features in the knowledge base. The obtained results indicate the overall accuracy of about 87% from the proposed multi-agent hyperspectral and Lidar decision fusion strategy. The obtained results from performing the same multi-agent system only on the hyperspectral image (without considering Lidar data) have the overall accuracy for about 71%.

The ability of recording the high resolution spectral signature of earth surface would be the most important feature of hyperspectral sensors. On the other hand, classification of hyperspectral imagery is known as one of the methods to extracting information from these remote sensing data sources. Despite the high potential of hyperspectral images in the information content point of view, there are a numerous challenges in reliable extraction of information from these images. The issues such as 1- spectral similarity of different phenomena, 2- sensor noises and atmospheric effects, 3- the effects of high dimensionality in the pattern recognition algorithms, 4- the necessity of large number of training data to perform a reliable classification, and 5- spectral variability of similar phenomena could be considered as some of the challenges in hyperspectral data processing. Decreasing of the high dimensionality effects via the dimension reduction algorithms (e.g. band selection and feature extraction algorithms), as well as increasing the separability of the overlapped classes through the linear/non-linear mappings into the feature spaces with the higher dimensional are two opposite and conventional approaches of hyperspectral data processing. These approaches would be used based on the factors such as 1- complexities of classes in the imaging area, 2- spectral range of imaging sensor, and 3- the restrictions of processing algorithms. In this paper the fusion of these two approaches is used to perform an accurate hyperspectral image classification. To do so, a novel feature extraction method is proposed to be used in the hyperspectral image classification. The core of this method is the fusion of the linear, non-linear and sparse representation based features which is used to produce the effective features in the weighted K-Nearest Neighbors (KNN) classification method. In this procedure, a set of supervised and nonlinear features are extracted as the first step through the Nonlinear Principal Component Analysis (NLPCA). The supervised usage of NLPCA in order to extract features is known as one of the novelties of this paper. In this step, the spectral bands are usually mapped to a high dimensional feature space through the self-estimator artificial neural networks (ANNs) which are trained separately by ground truth data. In the second step, the previously extracted features are linearly transformed by the Linear Discriminate Analysis (LDA) method in order to reduce the dimension of the hypercube generated via supervised NLPCA to a separable feature space. In the last step, a set of features which is proportional to the number of classes is generated based on the sparse representation theory. The sparse representation features were hired to handle the effects of the inter-class variability. The precisions of the classified features in the two different hyperspectral images were on average shown 6 percent improvements in comparison with the spectral bands and the other combinations of extracted features. Furthermore, reach to the approximately 99% overall accuracies in the classes with the few training data could be considered as other achievements of the proposed method.

High spectral resolution of hyperspectral images in form of very narrow and constant within visible and infrared spectral ranges has brought the technology of remote hyperspectral measurement into spotlight in order to detect objects as well as earthly phenomena. In this field, most methods have been presented with the purpose of improving the accuracy of image classification and there have been scarce researches on target detection (TD). Supervised  TD  problem  can  be  considered  as  a  one-class classification problem between the target and non-target pixels using  training  data  from  the  target  class  only.  However, a spectral signature of the target sample obtained using field or laboratory measurements is the only training data available for TD. A substantial number of bands lead to heavy computational costs along with Hughes Effect on hyperspectral image processing. Hence, in recent years much attention has been paid to reduction of computational complexity in the processing of hyperspectral images. In comparison to the classification  field,  few  studies  have  been  done  on  dimension  reduction  or  band  selection  for  target  detection  in  hyperspectral images. A chief reason behind this is the shortage or absence of training samples of the desired target in background images. In order to solve this problem, a method is introduced based on target simulation. Recently, target simulation method has been used for creating artificial sub-pixel targets on hyperspectral images in order to investigate the performance of sub-pixel target detection (STD) algorithms. But in this paper target simulation has been used as method for optimum band selection. In this method for STD several simulated training samples, created by means of target spectrum implantation in the image. For optimal band selection after achieving sufficient implanted targets as training data, searching strategy in hyperspectral image space is of high account. Once simulated targets are created, optimal bands are selected via Particle Swarm Optimization (PSO) Algorithm. To make the optimization algorithms exploit well in search space, their cost functions must be well defined. So one of contributions of this study is defining a new cost function for optimization algorithms used for selecting optimal bands. In the next stage, based on the optimal bands selected, the Local ACE is applied on the image to obtain detection result. ACE algorithm, one of the most useful and common algorithms for detection of sub-pixel and full pixel targets. But local ACE is commonly used to detect sub-pixel targets in hyperspectral images. In this version of the ACE algorithm, instead of using the mean and covariance of the entire image, just the neighbouring window pixels are used. The output of this stage will be TD map that by applying a threshold in a detection map can determine whether the pixels are target or not. In order to evaluate and study the ability of the introduced algorithm, Target Detection Blind Test (TDBT) of Hymap dataset and Hyperion dataset from Botswana have been used. False alarm rate was used as touchstone to evaluate the results between the outputs of different methods. Compared to some PSO-based algorithms such as maximum-submaximum-ratio (MSR) and correlation coefficient (CC), an evolutionary method such as genetic algorithm (GA), and the use of full band, the proposed method was able to improve the results by 46% in all cases where it was possible to decrease false alarm for the searched target.

Hyperspectral image contains hundreds of narrow and contiguous spectral bands. Because of this high spectral resolution, hyperspectral images provide valuable information from the earth surface materials and objects. By advances in remote sensing technology and production of hyper spectral data with high spatial and spectral information, using such data for a detailed study of the phenomenon is spreading quickly. One of the most important applications of hyperspectral data analysis is either supervised or unsupervised classification for land cover mapping. Among different unsupervised methods, Gaussian mixture model has attracted a lot of attention, due to its performance and efficient computational time. Gaussian Mixture Models (GMMs) have been frequently applied in hyperspectral image classification tasks. The problem of estimating the parameters in a Gaussian mixture model has been studied in the literature. Gibbs sampler is one of the methods that can be applied for this problem. Another method for estimation the parameters of a Gaussian mixture model is Expectation-Maximization (EM) algorithm. EM is a general method for optimizing likelihood functions and is useful in situations where data might be missing or simpler optimization methods fail .On the other hand, the large number of bands in a hyperspectral images leads into estimation of a large number of parameters. In the other point of view, the enormous amount of information provided by hyperspectral images increases the computational burden as well as the correlation among spectral bands. Thus, dimensionality reduction is often conducted as one of the most important steps before target detection to both maximize the detection performance and minimize the computational burden. In this paper, we use PCA and Random Projection (RP) for solving the high dimensionality of the data. In order to evaluate the proposed algorithm in real analysis scenarios, we used two benchmark hyperspectral data sets collected by AVIRIS and Reflective Optics System Spectrographic Imaging System (ROSIS). In order to evaluate the effectiveness of the proposed method which is based on the using GMMS and its parameter are estimated using Gibbs sampler method we used two well-known dataset ROSIS and AVIRIS hyperspectral images which they are acquired from a urban and agricultural area, respectively. Moreover, for better evaluation we used a simulated data which is attained using a toolbox which is known as HYDRA project. Investigations on the simulated dataset and two real hyperspectral data showed that the case in which the number of bands has been reduced in the pre-processing stage using either RP or PCA in the feature space, can result the highest accuracy and efficiency for thematic mapping. We also demonstrated that the superiority of the Gibbs sampler in comparison with EM algorithm for estimating the GMM parameters. For instance, in Pavia university dataset, the overall accuracy and Kappa coefficient was 88.80 and 0.84, respectively for GMM-Gibbs-RP method and for GMM-EM-RP method the overall accuracy and kappa coefficient was 84.21 and 0.80, respectively. In other view point, in urban area (Pavia university dataset) with small structures, the amount of improvement in by Gibbs sampler in comparison with EM algorithm was more than the AVIRIS dataset which is related to agricultural area with bigger regions. This shows the capability of Gibbs sampler in confronting with singularities.

Imaging spectroscopy, also known as hyperspectral imaging, is concerned with the measurement, analysis, and interpretation of spectra acquired from either a given scene or a specific object at a short, medium, or long distance by a satellite sensor over the visible to infrared and sometime thermal spectral regions. The recent developments in spatial, spectral and radiometric resolution of hyperspectral images have stimulated new methodologies for land cover and land use classification. There are two major approaches for classification of hyperspectral images: the spectral or pixel-based and the spectral-spatial or object-based approaches. While the pixel-based techniques use only the spectral information of the pixels, the spectral-spatial frameworks employ both spectral characteristics and spatial context of the pixels. The pixel-based classification methods are often unable to accurately differentiate between some classes with high spectral similarity. This is mainly because they employ only the spectral information in order to identify different land covers. Consequently, methods that can exploit the spatial information are essential for more accurate classification results.
Among the various methods for extracting spatial information, segmentation techniques are the powerful tools for defining the spatial dependences among the pixels and finding the homogeneous regions in the image. An alternative way to achieve the accurate segmentations of image is marker-controlled segmentation. The idea behind this approach is selecting of one or several pixels for every spatial object as the seed or a marker of the corresponding region. The marker-based segmentation significantly reduced the over-segmentation problem and led to better accuracy rate. Recently, an effective approach to spectral-spatial classification of hyperspectral images has been proposed based on Minimum Spanning Forest (MSF) grown from automatically selected markers using Support Vector Machines (SVM) classification. In this framework, a connected components labelling is applied on the classification map. Then, if a region is large enough, its marker is determined as the P% of pixels within this region with the highest probability estimates. Otherwise, it should lead to a marker only if it is very reliable. A potential marker is formed by pixels with estimated probability higher than a defined threshold.
This paper aims at improving this approach by reducing the spatial dimensions of hyperspectral images. The proposed approach are evaluated the dimension reduction of hyperspectral image before and after marker selection process in MSF using genetic algorithm. The genetic algorithm is a general adaptive optimization search method based on a direct analogy to Darwinian natural selection and genetics in biological systems. It starts from an initial population which is composed of a set of possible solutions called individuals (chromosomes), and then evaluates the quality of each individual based on a fitness function. We use the Kappa coefficient accuracy parameter of SVM classification obtained from the training samples subset as the fitness function. Three benchmark hyperspectral datasets are used for evaluation: the Pavia dataset, the Telops dataset and the Indian Pines dataset. Experimental results show the superiority of using genetic algorithm before selecting markers in Pavia and Telops datasets. In Indian Pines dataset, the classification accuracy was increased with reduced dimensions both before and after the marker selection and concurrently.

Hyperspectral images، with high spectral resolution، have caused vast progress in remote sensing extensions. One of the most important applications of these images is agriculture and forest. The purpose of this research is improvement in classification of various vegetation types over Botswana region by using combination of target detection algorithms and Hyperspectral image. In the first step، target detection algorithms implemented over the preprocessed Hyperspectral image. In the second step، information of target detection algorithms was combined by using the proposed method. Results of the proposed method were implemented for different windows size. The best overall accuracy of the method was 96. 16 percent for 3*3 windows size that its overall accuracy has approximately improved at least 8 percent and uttermost 20 percent with respect to the results of target detection algorithms.

Recently, hyperspectral images analysis has obtained successful results from information extraction in urban areas. Building detection is one of the important applications in processing hyperspectral images. In order to detect complete and precise building information from hyperspectral data, advanced data analysis methods are required. Algorithms based on spectral-identification are sensitive to spectral variability and noise in acquisition. In most cases, the spectral signature is unknown, so each pixel is separately examined and if it significantly differs from the background, it is regarded as an object. On the other hand, there are many algorithms e.g. Spectral Angle Measure (SAS), Spectral Correlation Similarity (SCS), Spectral Information Divergence (SID), Jeffries-Matusita Distance (JMD), Constrained Energy Minimizing (CEM) and Covariance-based Matched Filter Measure (CMFM) for building detection. In this study, first we employed the SAS, SCS, SID, JMD, CEM and CMFM algorithms for building detection. Then, in the next step to improve the spectral detection algorithms, two strategies, the combining algorithms using Adaptive Neuro-Fuzzy Inference System (ANFIS) method and spectral-spatial detection, was employed. Our experiments results demonstrate a significant improvement of accuracy using proposed strategies on two CASI hyperspectral images taken from an urban area.