Shouda Jiang

Automatic Target Detection and Tracking in FLIR Image Sequences Using Morphological Connected Operator

In this paper, we propose a method for detecting and tracking small targets in forward looking infrared (FLIR) image sequences taken from an airborne moving platform. Firstly, we adopt the morphological connected operator to remove the undesirable clutter in the background. Secondly, the image is decomposed by morphological Haar wavelet, and the wavelet energy image is computed from the horizontal and vertical detail images, and it is fused with the scaled image. Thirdly, the targets are extracted coarse-to-fine by adaptive double thresholding. Finally, targets are modeled by intensity probabilistic density function and tracked using mean shift algorithm. The experiments performed on the AMCOM FLIR data set verify the validity and robustness of the algorithm.

Hierarchical search strategy in particle filter framework to track infrared target

A target of interest may exhibit significant appearance variations because of its complex maneuvers, ego-motion of the camera platform, etc. Currently, target tracking in forward-looking infrared (FLIR) sequences is still a challenging problem in the field of computer vision. Although many efforts have been devoted, there are still some issues to be addressed. First, state particles generated by prior information cannot approximate the probability density function well when the target state changes obviously. Second, plenty of particles have to be employed to obtain satisfying estimation of target state which will cause heavy computational burden in turn. In this paper, a hierarchical search strategy (HS tracker) is proposed to track infrared target in the particle filter framework, and there are two observation models employed to locate the target robustly. In the first stage, a saliency map leads the redistributed state particles to cover the salient areas that can provide a rough prediction of the target areas. In the second stage, sparse representation is employed to search for a subset of true ones from all the target candidates; thus, only efficient state particles are used to estimate the target state. The proposed method is tested on numerous FLIR sequences from the US army aviation and missile command database, and experimental results demonstrate the excellent performance.

A dual-layer supervised Mahalanobis kernel for the classification of hyperspectral images

To address the drawback of traditional Mahalanobis distance metric learning (DML) methods that learn the matrix without considering the weights of each class, in this paper, a novel dual-layer supervised Mahalanobis kernel is proposed for the classification of hyperspectral images. By modifying the traditional unsupervised Mahalanobis kernel, a supervised Mahalanobis matrix that can include more relativity information of different types of real materials in hyperspectral images is learned to obtain a new kernel. The proposed Mahalanobis matrix is obtained in two steps. In step one, we learn the first traditional Mahalanobis matrix with all samples to map the raw data. In step two, based on the data mapped by the first matrix, we pick several hard-to-identify classes from all the classes and learn the second Mahalanobis matrix using only these data. Finally, by combining these two matrices, we construct a new form of the Mahalanobis kernel. Simulation experiments are conducted on three real hyperspectral data sets. We use SVM as the kernel-based classifier to classify the dimensionally reduced data and compare with several methods from various aspects. The results show that the proposed methods perform better than other unsupervised or single-layer DML methods in classifying the hard-to-identify classes, especially under an extreme condition.

A Fast Training Algorithm for Least Squares SVM

A fast training algorithm for Least Squares SVM (LS-SVM) classifiers was proposed, which is based on incremental and decremental learning theory. When a SV (Support Vector) is added or removed, computation based on previous training result replaces large-scale matrix inverse, thus the computation cost is reduced. The innovation is that by reasonable use of incremental and decremental learning the proposed algorithm can adaptively adjust the size of training sets (number of SVs) according to the specific classification problem. Finally several experiments show the validity of proposed algorithm.

A Mahalanobis metric learning-based polynomial kernel for classification of hyperspectral images

In this paper, to combine the advantage of both polynomial kernel and the Mahalanobis distance metric learning (DML) methods, we propose a Mahalanobis DML based polynomial kernel for the classification of hyperspectral images. To ensure the method is computing-saving, we adapt a fast iterative method to learn the Mahalanobis matrix. Simulation experiment is conducted on two real hyperspectral data sets. To evaluate the proposed method, we compare it with the traditional radial basis function (RBF) kernel, polynomial kernel and the RBF-based Mahalanobis kernel, the result shows the performance of the proposed method did improve the capability of the polynomial kernel and also perform better than the RBF-based Mahalanobis kernel.

A dual-kernel spectral-spatial classification approach for hyperspectral images based on Mahalanobis distance metric learning

Abstract Hyperspectral images provide a precise representation of the earth’s surface, with abundant spectral and spatial features, but normal classification algorithms use only the information provided by the spectral features of each data point. In this paper, we propose a new approach to hyperspectral image classification based on Mahalanobis distance metric learning and kernel learning that considers both the features of the spectral bands and a spatial prior. This approach consists of two components. First, we obtain a primary labeled classification result and a posterior probability distribution for each pixel point using a Mahalanobis-kernel-based classifier. Second, instead of the original or extracted spectral features, we reconstruct the spatial relationship of the hyperspectral images using the posterior probability of every data point, smooth the boundaries, and revise suspicious points based on this piecewise information using a kernel-based multi-region segmentation method. In an experimental study, we adopt a support vector machine (SVM) classifier as the kernel classifier to obtain the posterior probabilities using dimensionally reduced data. The proposed method is compared with several other methods from various perspectives. Simulation experiments run on several real hyperspectral data sets are reported. The results show that the proposed method performs better than other comparable classification algorithms, especially in a condition-constrained environment.

An Infrared Small Target Detection Method Based on Block Compressed Sensing

Aiming at improving the real-time performance of infrared weapon systems, a method based on block compressed sensing is proposed to detect infrared small target, which is also easy to be implemented with hardware. The proposed method can detect and locate small infrared targets by classifying compressed results of blocks. In addition, in order to solve the low detection accuracy caused by using uniform block, the proposed method uses overlapping blocks to reduce the maximum distance between the center of the test sample block and that of the target. Experiments show that the proposed method can effectively improve the detection accuracy of infrared small targets.

Computation of Large-Scale Electric Field in Free Space

A method of modeling Spatial Electric Field around the earth is presented to provide the spatial electric field data in virtual test system. Firstly, the electric field in different time in the condition of sunny day is modeled to generate the electric field data in a period time and spatial. Then, this paper gives the analysis of the influence to Spatial Electric Field of thunderstorm.