Saeid Homayouni

Anomaly Detection in Hyperspectral Images Based on an Adaptive Support Vector Method

Recently, anomaly detection (AD) has attracted considerable interest in a wide variety of hyperspectral remote sensing applications. The goal of this unsupervised technique of target detection is to identify the pixels with significantly different spectral signatures from the neighboring background. Kernel methods, such as kernel-based support vector data description (SVDD) (K-SVDD), have been presented as the successful approach to AD problems. The most commonly used kernel is the Gaussian kernel function. The main problem using the Gaussian kernel-based AD methods is the optimal setting of sigma. In an attempt to address this problem, this paper proposes a direct and adaptive measure for Gaussian K-SVDD (GK-SVDD). The proposed measure is based on a geometric interpretation of the GK-SVDD. Experimental results are presented on real and synthetically implanted targets of the target detection blind-test data sets. Compared to previous measures, the results demonstrate better performance, particularly for subpixel anomalies.

An Approach for Subpixel Anomaly Detection in Hyperspectral Images

Fast detecting difficult targets such as subpixel objects is a fundamental challenge for anomaly detection (AD) in hyperspectral images. In an attempt to solve this problem, this paper presents a novel but simple approach based on selecting a single feature for which the anomaly value is the maximum. The proposed approach applied in the original feature space has been evaluated and compared with relevant state-of-the-art AD methods on Target Detection Blind Test data sets. Preliminary results suggest that the proposed method can achieve better detection performance than its counterparts. The results also show that the proposed method is computationally expedient.

Hyperspectral image analysis for precision viticulture

We analyze the capabilities of CASI data for the discrimination of vine varieties in hyperspectral images. To analyze the discrimination capabilities of the CASI data, principal components analysis and linear discriminant analysis methods are used. We assess the performance of various classification techniques: Multi-layer perceptrons, radial basis function neural networks, and support vector machines. We also discuss the trade-off between spatial and spectral resolutions in the framework of precision viticulture.

An Improved FCM Algorithm Based on the SVDD for Unsupervised Hyperspectral Data Classification

Unsupervised classification approaches, also known as “clustering algorithms”, can be considered a solution to problems associated with the supervised classification of remotely sensed image data. The most important of these problems with respect to statistical classification algorithms is the lack of enough high quality training data and high dimensionality of hyperspectral data. In this paper, an improved clustering framework is developed and evaluated as a resolution to these problems. The proposed method enhances the Fuzzy C-Means (FCM) algorithm by using the Support Vector Domain Description (SVDD). The proposed algorithm operates in a similar manner as the FCM for the clustering and labeling of data vectors. However, for estimation of the cluster centers, the SVDD encircles the corresponding members and estimates the center of a containing sphere. By doing so, the effects of noise and outliers on the cluster centers are reduced, and more specifically, higher classification accuracy can be obtained. In spite of this advantage, there are two sets of parameters, namely, the SVDDs and FCMs parameters, both of which affect the performance of the proposed algorithm. Accordingly, the effects of these parameters and their optimum values have been evaluated as well. The evaluations of the results of experiments show that the proposed algorithm, due to the use of the SVDD algorithm, is more efficient than other clustering algorithms.

Improving the SVDD Approach to Hyperspectral Image Classification

In recent studies, the support vector data description (SVDD) has been successfully applied to the classification of hyperspectral images. However, there is a major problem with this approach, namely, the precise setting of the Gaussian kernel width (i.e., the sigma), which is, in fact, the common limitation of kernel methods in achieving a reliable performance. Generally, the sigma is tuned for multiclass data sets through the K-fold cross validation (K CV), a time-consuming method. To reduce the computation time in real-time applications, typically, the KCV is used to constrain all the involved SVDD classifiers to share the same sigma. This letter presents a fast and straightforward method to estimate the sigma for each individual SVDD classifier based on statistical properties of the Gaussian kernel. To evaluate the performance of the proposed method, three frequently used hyperspectral data sets are employed. The results are then compared to the KCV method for sigma selection, and, in addition, two direct sigma estimation methods. Preliminary results using incomplete training data suggest that the proposed method can achieve similar or better performance with faster processing times than the KCV and also provide a significant superior performance in comparison with the direct methods.

Range Camera Self-Calibration Based on Integrated Bundle Adjustment via Joint Setup with a 2D Digital Camera

Time-of-flight cameras, based on Photonic Mixer Device (PMD) technology, are capable of measuring distances to objects at high frame rates, however, the measured ranges and the intensity data contain systematic errors that need to be corrected. In this paper, a new integrated range camera self-calibration method via joint setup with a digital (RGB) camera is presented. This method can simultaneously estimate the systematic range error parameters as well as the interior and external orientation parameters of the camera. The calibration approach is based on photogrammetric bundle adjustment of observation equations originating from collinearity condition and a range errors model. Addition of a digital camera to the calibration process overcomes the limitations of small field of view and low pixel resolution of the range camera. The tests are performed on a dataset captured by a PMD[vision]-O3 camera from a multi-resolution test field of high contrast targets. An average improvement of 83% in RMS of range error and 72% in RMS of coordinate residual, over that achieved with basic calibration, was realized in an independent accuracy assessment. Our proposed calibration method also achieved 25% and 36% improvement on RMS of range error and coordinate residual, respectively, over that obtained by integrated calibration of the single PMD camera.

Classification of polarimetric SAR images using Support Vector Machines

Recently, Support Vector Machines (SVMs) have been introduced as a promising tool for performing supervised classification. This approach has been applied in different contexts and applications, such as data mining, regression analysis, and the classification of remotely sensed data. The advantage of SVMs for data classification is their ability to be used as an efficient algorithm for nonlinear classification problems, particularly in the case of extracting feature vectors from fully polarimetric SAR data. In this research, a classification algorithm based on the SVMs technique is applied to the fully polarimetric AIRSAR L-band data from the San Francisco Bay area, with a spatial resolution of 10 m. Several parameters are extracted from SAR data, including the individual channel backscatter value, Pauli decomposition coefficients, Krogager decomposition coefficients, and eigenvector decomposition parameters. Different combinations of polarimetric parameters are considered to assess the accuracy of the classification results. The accuracy of the SVMs is then compared with that obtained from several conventional classifiers, including the Maximum Likelihood classifier, Minimum Distance classifier, Mahalanobis Distance classifier, and Wishart classifier. The accuracy analysis shows that, for classification of fully polarimetric data, SVMs perform more poorly than the Wishart classifier by approximately 16%, whereas they perform better than the Maximum Likelihood, Minimum Distance, and Mahalanobis Distance classifiers by approximately 4%, 17% and 14%, respectively. Moreover, the highest accuracy is achieved by using the coefficients of Krogager decomposition in the classification procedure. This evaluation demonstrates that the SVM classifier can be used as an effective method for analyzing fully polarimetric SAR images with acceptable levels of accuracy.

A Hybrid Kernel-Based Change Detection Method for Remotely Sensed Data in a Similarity Space

Detection of damages caused by natural disasters is a delicate and difficult task due to the time constraints imposed by emergency situations. Therefore, an automatic Change Detection (CD) algorithm, with less user interaction, is always very interesting and helpful. So far, there is no existing CD approach that is optimal and applicable in the case of (a) labeled samples not existing in the study area; (b) multi-temporal images being corrupted by either noise or non-normalized radiometric differences; (c) difference images having overlapped change and no-change classes that are non-linearly separable from each other. Also, a low degree of automation is not optimal for real-time CD applications and also one-dimensional representations of classical CD methods hide the useful information in multi-temporal images. In order to resolve these problems, two automatic kernel-based CD algorithms (KCD) were proposed based on kernel clustering and support vector data description (SVDD) algorithms in high dimensional Hilbert space. In this paper (a( a new similarity space was proposed in order to increase the separation between change and no-change classes, and also to decrease the processing time, (b) three kernel-based approaches were proposed for transferring the multi-temporal images from spectral space into high dimensional Hilbert space, (c) automatic approach was proposed to extract the precise labeled samples; (d) kernel parameter was selected automatically by optimizing an improved cost function and (e) initial value of the kernel parameter was estimated by a statistical method based on the L2-norm distance. Two different datasets including Quickbird and Landsat TM/ETM+ imageries were used for the accuracy of analysis of proposed methods. The comparative analysis showed the accuracy improvements of kernel clustering based CD and SVDD based CD methods with respect to the conventional CD techniques such as Minimum Noise Fraction, Independent Component Analysis, Spectral Angle Mapper, Simple Image differencing and Image Rationing, and also the computational cost analysis showed that implementation of the proposed CD method in similarity space decreases the processing runtime.

A SVMS-based hyperspectral data classification algorithm in a similarity space

In this paper a multi-steps algorithm based on Support Vectors Machines (SVMs) in similarity space is proposed. The SVMs is used as a recent classification method and separation boundary estimation technique for high dimensional data. It benefits of limited number of data for training of supervised classification, which is a key challenge in hyperspectral data analysis.

Abundance weighting for improved vegetation mapping in row crops: application to vineyard vigour monitoring

We present a complete framework for vigour mapping in row crops by multispectral remote sensing. The main contribution consists of taking into account vegetation abundance in the computation of vigour indexes. Though developed in a viticulture context, the proposed algorithm is generic enough to be adapted to any row crop, especially in horticulture. The algorithm takes advantage of both spectral and spatial features extracted from image data. Spectral information is used at pixel level by an independent component analysis (ICA) based algorithm to process vegetation abundance maps. As for spatial information, deformable models are used to fit a network of rectangles to individual plants. Both spectral information and spatial information are then combined to compute abundance-weighted vigour indexes that are assigned to specific plants. Resulting measurements are then used for within-block vigour mapping. A validation procedure is carried out on experimental vine plots. It is shown that the use of vegetation abundance by itself or as a weight in the computation of vegetation indexes improves the accuracy of vigour assessment in row crops.