filter by:
Articles
Publication Date: 2025
IEEE Transactions on Geoscience and Remote Sensing (15580644)63
In this article, a multimodal deep architecture for classification of light detection and ranging (LiDAR) and hyperspectral image (HSI) is proposed, acquiring the knowledge of both modalities by leveraging modality-specific information and their complementary information. The proposed model consists of two main steps. First, to improve the performance of a 2-D convolutional neural network (2DCNN), low-frequency features with a maximum autocorrelation factor of HSI are injected into 2DCNN, which are called multiscale features of 2DCNN. Second, to improve the accuracy of 2DCNN and extract smooth and semantic information, the posterior energy of hidden Markov random field (HMRF) is modified by using Gaussian attention and albedo recovery attention mechanisms and energies of LiDAR and HMRF. Then, these features are fused based on another attention mechanism called attention-based HMRF. Moreover, this HMRF model is used for the fusion of HSI and LiDAR. The proposed model is tested on the Houston 2013, Trento, and MUUFL datasets and compared with several state-of-the-art methods. The resulting classification accuracies through the ablation study show the superior performance of the proposed method. © 1980-2012 IEEE.
Publication Date: 2021
Signal Processing: Image Communication (09235965)92
This article proposes a spectral–spatial method for classification of hyperspectral images (HSIs) by modifying traditional Auto-Encoder based on Majorization Minimization (MM) technique. The proposed method consists of suggesting three main modifications. First, to construct weights of Auto-Encoder, similarity angle map(SAM) criterion is used as regularization term. It is useful to extract spectral similarity of initial features. Second, to enhance the classification accuracy, fuzzy mode is used to estimate parameters. These modifications lead to create an extended Auto-Encoder based on MM (EAEMM). Third, to improve the performance of Auto-Encoder, multi-scale features (MSF) are extracted. In comparison with some of the state-of-the-art methods, the experimental results obtained using the proposed method (MSF-EAEMM) show that it significantly improves the classification accuracy of HSI classification. © 2020
Publication Date: 2019
International Journal of Remote Sensing (13665901)40(10)pp. 3948-3967
This article proposes a spectral–spatial method for classification of hyperspectral images (HSIs) by modifying traditional random walker (RW). The proposed method consists of suggesting two main modifications. First, to construct a spatial edge weighting function, low-frequency edge weighting function is proposed. In this function, the detail weights are removed. Second, to enhance the classification accuracy, a fusion of spectral and spatial Laplacian matrix in RW is suggested. This fusion can improve the classification performances compared to traditional RW using only spatial Laplacian matrix. In comparison with some of the state-of-the-art RW and spectral–spatial classifier methods, the experimental results of the proposed method (spectral–spatial RW) show that the proposed method significantly increases the classification accuracy of HSI. © 2018, © 2018 Informa UK Limited, trading as Taylor & Francis Group.
Publication Date: 2018
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing (21511535)11(11)pp. 4101-4112
This paper proposes a spectral-spatial method for the classification of hyperspectral images (HSIs). The proposed method consists of suggesting two main techniques. First, to construct dictionaries and sparse codes containing the least spatial-spectral correlation, an objective function for a sparse classifier, namely optimum dictionary learning (ODL), is proposed. This objective function combines the spectral and spatial features in a hierarchical fashion. ODL is applied in the nonsubsampled pyramid domain to exploit the multiresolution property of pyramids. Second, to enhance the classification accuracy, an extended hidden Markov random field (EHMRF) is suggested. The EHMRF uses the Poisson distribution and weighted mean regularization to overcome Poisson noise and to increase the local neighborhood energy consistency, respectively. Finally, in order to estimate the parameter of the Poisson distribution, an iterative maximum likelihood (ML) algorithm is used. In comparison with some of state-of-the-art hidden Markov random field and sparse-based classifier methods, the experimental results of the proposed method (ODL-EHMRF-ML) show that it significantly increases the classification accuracy of HSIs. © 2018 IEEE.
Publication Date: 2018
Geocarto International (17520762)33(8)pp. 771-790
This paper proposes a spectral–spatial method for classification of hyperspectral images. The proposed method, called SSC, consists of two steps. In the first step, to overcome the computation complexity, a wavelet-based classifier is designed. In the second step, to enhance the classification accuracy, a novel hidden Markov random field called NHMRF technique in spatial domain is suggested. In NHMRF, we convert two-dimensional energies of traditional hidden Markov random field to three-dimensional energies and then we apply edge preserving regularization terms on each two-dimensional energy of this cube. The class label of each test pixel is fixed based on minimum three-dimensional energy achieved by edge preserving regularization terms. Experimental results show that the classification accuracy of the proposed approach based on three-dimensional energies and edge preserving regularization terms is effectively improved in comparison with the state-of-the-art methods. © 2017 Informa UK Limited, trading as Taylor & Francis Group.
Publication Date: 2017
International Journal of Remote Sensing (13665901)38(12)pp. 3608-3634
This article proposes a new algorithm for hyperspectral image classification. The proposed method is a spectral–spatial method based on wavelet transforms, kernel minimum noise fraction (KMNF) and spatial–spectral Schroedinger eigenmaps (SSSE). To overcome the computation complexity, one-dimensional discrete wavelet transform (1D-DWT) is applied in spectral domain. To reduce noise, KMNF coefficients are extracted in wavelet space. To solve time-consuming problem, 2D-DWT coefficients are employed in spatial space. Hence, the combination of 1D-DWT, KMNF, and 2D-DWT is suggested to create SSSE features. The classification is carried out by a Support Vector Machine (SVM) classifier. Experimental results show that classification accuracy and time consumption are effectively improved compared to the state-of-the art reported spectral–spatial SVM-based methods. © 2017 Informa UK Limited, trading as Taylor & Francis Group.
Publication Date: 2017
Remote Sensing (20724292)9(6)
In this paper, a new method is presented for spatial resolution enhancement of hyperspectral images (HSI) using spectral unmixing and a Bayesian sparse representation. The proposed method combines the high spectral resolution fromtheHSIwith the high spatial resolution fromamultispectral image (MSI) of the same scene and high resolution images fromunrelated scenes. The fusionmethod is based on a spectral unmixing procedure for which the endmember matrix and the abundance fractions are estimated from the HSI and MSI, respectively. A Bayesian formulation of this method leads to an ill-posed fusion problem. A sparse representation regularization term is added to convert it into a well-posed inverse problem. In the sparse representation, dictionaries are constructed from the MSI, high optical resolution images, synthetic aperture radar (SAR) or combinations of them. The proposed algorithm is applied to real datasets and compared with state-of-the-art fusion algorithms based on spectral unmixing and sparse representation, respectively. The proposed method significantly increases the spatial resolution and decreases the spectral distortion efficiently. © 2017 by the authors.
Publication Date: 2012
pp. 616-620
In this paper, a method for the fusion of IR and color images is proposed which can use both Non-Subsampled Contourlet Transform (NSCT) and combination of Haar wavelet transform and Unsharp filtering. We fuse images with combination of low level Haar Wavelet and Unsharp filtering instead of the input images. Moreover, image fusion is done in bandpass and lowpass subbands, which are obtained by NSCT. This work has three advantages over the use of single NSCT. Firstly the elapsed time decreases. Secondly, high frequency coefficients (i.e. high freq. wavelet images and bandpass subband NSCT images) can be fused twice. Finally, by applying unsharp filtering, the quality of reconstructed Wavelet image is enhanced and features based on combining Wavelet and NSCT are more effectively used. The obtained results show a better performance of the proposed method. © 2012 Institute of Telecommunica.
In this paper, we propose an algorithm of license plates recognition from their images captured by a camera in digital zoom using a binary time delay neural network (TDNN). Moreover, hard conditions such as the distance and angle variations as well as weather and light conditions are considered. For training the neural network, we collected the training images using the Zernike moment when the camera was not in the magnification state and the test images when the camera was in zooming state. The comparison was made between the proposed algorithm and the previous methods in character recognition like SVM and classical TDNN. The algorithms have been evaluated using 50 license plate images with magnification of 8. The recognition rate obtained by the proposed algorithm was 70%. © 2010 IEEE.
