Search

Satellite image processing has been widely used in recent years in a number of applications such as land classification, Identification transfer, resource exploration, super‐resolution image, etc. Due to the orbital location, revision time, quick view angle limitations, and weather impact, the satellite images are challenging to manage. There are many types of resolution, such as spatial, spectral, and temporal. Still, in our case, we concentrated on spatial image resolution to super resolve the images from low‐resolution images. For remote sensing image super‐resolution fast wavelet‐based super‐resolution (FWSR), we propose a novel, fast wavelet‐based plexus framework that performs super‐resolution convolutional neural network (SRCNN)‐like extraction of features based on three hidden layers. First, wavelet sub‐band images are combined into a pre‐defined full‐scale data training factor, including approximation and interchangeable stand‐alone units (frequency sub‐bands). Second, to speed up image recovery, mapping the sub‐band image of the wavelet is then measured using its approximate image. Third, the added sub‐pixel layer at the end of the network model is intended to reproduce image quality using a plexus framework. The approximation sub‐band images obtained after discrete wavelet transform wavelet decomposition are used as input rather than the original image because of their high‐frequency data and preserved characteristics. Five current super‐resolution neural network approaches are compared with the proposed technique and tested on three pubic satellite image datasets and two benchmark datasets. The experimental findings are well compared qualitatively and quantitatively.

Super-resolution (SR) has received extensive attention in recent years for satellite image processing in a wide range of application scenarios, such as land classification, identification of changes, the discovery of resources, etc. Satellite images from satellite sensors are mostly low-resolution (LR) images, so they do not completely fulfill object detection and analysis criteria. SR has multiple residual network frameworks in deep learning that have improved performance and can extend thousands of layers in the system. However, each layer improves accuracy by doubling the number of layers, although training thousands of layers are too expensive, the process is slow, and there are functional recovery issues. We proposed a transferred wide residual Single Image Super-Resolution (SISR) remote sensing deep neural network model (WRSR). By increasing the width and reducing the residual network depth, the proposed approach has dramatically reduced memory costs. As a result, our model reduced memory costs by 21% in Enhanced Deep Residual Super-Resolution (EDSR) and 34% in SRResNet as a direct consequence of the in-depth reduction. The proposed architecture improves the efficiency of training loss by performing weight normalization instead of augmentation technology. We compared our method to five recent existing super-resolution (SR) deep neural network methods, tested over three public satellite image datasets and a standard reference (PRIM) dataset. Experiment analysis is evaluated in peak to signal noise ratio (PSNR) and structural similarity index measure (SSIM).

The object identification within an image captured during rough weather conditions (such as haze, fog) poses difficulty due to the reduction of an image. The rough weather conditions lead not only to the variation of the image's visual effect but also to the disadvantage of post‐processing of an image. Furthermore, it causes inconvenience of all types of instruments that rely on optical imaging, such as satellite remote‐sensing systems, aerial photo systems, outdoor monitoring systems, and object identification systems, respectively. Hence, the improvement and restorement of the visual effects and enhanced post‐processing are needed. This research introduces a new image enhancement approach for image dehazing based on dark channel prior and piecewise linear transformation; also, the histogram equalisation technique, i.e. contrast limited adaptive histogram equalisation is applied. A dark channel prior is well known for its simplicity and productivity. In this work, the dark channel prior to a new angle is analysed in the first step, where average patch sizes are estimated for the computation of haze densities. Furthermore, the sky is approximated up to 5–10% of the hazy images, which has a good effect in removing the haze from the image. Using the dark channel, the proposed algorithm significantly boosted the effects of the dark images as well as reduced the influence of haze and noise. Eventually, for colour correction, the piecewise linear transformation technique is applied, which enhances the colour close to the original image. Experimental results demonstrate that the proposed method significantly improves the visibility of the algorithm on dark remote‐sensing images as well as on hazy natural images.

Absolute localization of a flying UAV on its own in a global-navigation-satellite-system (GNSS)-denied environment is always a challenge. In this paper, we present a landmark-based approach where a UAV is automatically locked into the landmark scene shown in a georeferenced image via a feedback control loop, which is driven by the output of an aerial image registration. To pursue a real-time application, we design and implement a speeded-up-robust-features (SURF)-based image registration algorithm that focuses efficiency and robustness under a 2D geometric transformation. A linear UAV controller with signals of four degrees of freedom is derived from the estimated transformation matrix. The approach is validated in a virtual simulation environment, with experimental results demonstrating the effectiveness and robustness of the proposed UAV self-localization system.

Training and evaluation of fair classifiers is a challenging problem. This is partly due to the fact that most fairness metrics of interest depend on both the sensitive attribute information and label information of the data points. In many scenarios it is not possible to collect large datasets with such information. An alternate approach that is commonly used is to separately train an attribute classifier on data with sensitive attribute information, and then use it later in the ML pipeline to evaluate the bias of a given classifier. While such decoupling helps alleviate the problem of demographic scarcity, it raises several natural questions such as: how should the attribute classifier be trained?, and how should one use a given attribute classifier for accurate bias estimation? In this work we study this question from both theoretical and empirical perspectives. We first experimentally demonstrate that the test accuracy of the attribute classifier is not always correlated with its effectiveness in bias estimation for a downstream model. In order to further investigate this phenomenon, we analyze an idealized theoretical model and characterize the structure of the optimal classifier. Our analysis has surprising and counter-intuitive implications where in certain regimes one might want to distribute the error of the attribute classifier as unevenly as possible among the different subgroups. Based on our analysis we develop heuristics for both training and using attribute classifiers for bias estimation in the data scarce regime. We empirically demonstrate the effectiveness of our approach on real and simulated data.

Continual learning has become increasingly important as it enables NLP models to constantly learn and gain knowledge over time. Previous continual learning methods are mainly designed to preserve knowledge from previous tasks, without much emphasis on how to well generalize models to new tasks. In this work, we propose an information disentanglement based regularization method for continual learning on text classification. Our proposed method first disentangles text hidden spaces into representations that are generic to all tasks and representations specific to each individual task, and further regularizes these representations differently to better constrain the knowledge required to generalize. We also introduce two simple auxiliary tasks: next sentence prediction and task-id prediction, for learning better generic and specific representation spaces. Experiments conducted on large-scale benchmarks demonstrate the effectiveness of our method in continual text classification tasks with various sequences and lengths over state-of-the-art baselines. We have publicly released our code at https://github.com/GT-SALT/IDBR., NAACL 2021

As multi-task models gain popularity in a wider range of machine learning applications, it is becoming increasingly important for practitioners to understand the fairness implications associated with those models. Most existing fairness literature focuses on learning a single task more fairly, while how ML fairness interacts with multiple tasks in the joint learning setting is largely under-explored. In this paper, we are concerned with how group fairness (e.g., equal opportunity, equalized odds) as an ML fairness concept plays out in the multi-task scenario. In multi-task learning, several tasks are learned jointly to exploit task correlations for a more efficient inductive transfer. This presents a multi-dimensional Pareto frontier on (1) the trade-off between group fairness and accuracy with respect to each task, as well as (2) the trade-offs across multiple tasks. We aim to provide a deeper understanding on how group fairness interacts with accuracy in multi-task learning, and we show that traditional approaches that mainly focus on optimizing the Pareto frontier of multi-task accuracy might not perform well on fairness goals. We propose a new set of metrics to better capture the multi-dimensional Pareto frontier of fairness-accuracy trade-offs uniquely presented in a multi-task learning setting. We further propose a Multi-Task-Aware Fairness (MTA-F) approach to improve fairness in multi-task learning. Experiments on several real-world datasets demonstrate the effectiveness of our proposed approach.

Single Image Super-resolution (SISR) produces high-resolution images with fine spatial resolutions from aremotely sensed image with low spatial resolution. Recently, deep learning and generative adversarial networks(GANs) have made breakthroughs for the challenging task of single image super-resolution (SISR). However, thegenerated image still suffers from undesirable artifacts such as, the absence of texture-feature representationand high-frequency information. We propose a frequency domain-based spatio-temporal remote sensingsingle image super-resolution technique to reconstruct the HR image combined with generative adversarialnetworks (GANs) on various frequency bands (TWIST-GAN). We have introduced a new method incorporatingWavelet Transform (WT) characteristics and transferred generative adversarial network. The LR image hasbeen split into various frequency bands by using the WT, whereas, the transfer generative adversarial networkpredicts high-frequency components via a proposed architecture. Finally, the inverse transfer of waveletsproduces a reconstructed image with super-resolution. The model is first trained on an external DIV2 Kdataset and validated with the UC Merceed Landsat remote sensing dataset and Set14 with each image sizeof 256x256. Following that, transferred GANs are used to process spatio-temporal remote sensing images inorder to minimize computation cost differences and improve texture information. The findings are comparedqualitatively and qualitatively with the current state-of-art approaches. In addition, we saved about 43% of theGPU memory during training and accelerated the execution of our simplified version by eliminating batchnormalization layers., Comment: Accepted: ACM TIST (10-03-2021)

We present ToTTo, an open-domain English table-to-text dataset with over 120,000 training examples that proposes a controlled generation task: given a Wikipedia table and a set of highlighted table cells, produce a one-sentence description. To obtain generated targets that are natural but also faithful to the source table, we introduce a dataset construction process where annotators directly revise existing candidate sentences from Wikipedia. We present systematic analyses of our dataset and annotation process as well as results achieved by several state-of-the-art baselines. While usually fluent, existing methods often hallucinate phrases that are not supported by the table, suggesting that this dataset can serve as a useful research benchmark for high-precision conditional text generation., Accepted to EMNLP 2020

NLP models are shown to suffer from robustness issues, i.e., a model's prediction can be easily changed under small perturbations to the input. In this work, we present a Controlled Adversarial Text Generation (CAT-Gen) model that, given an input text, generates adversarial texts through controllable attributes that are known to be invariant to task labels. For example, in order to attack a model for sentiment classification over product reviews, we can use the product categories as the controllable attribute which would not change the sentiment of the reviews. Experiments on real-world NLP datasets demonstrate that our method can generate more diverse and fluent adversarial texts, compared to many existing adversarial text generation approaches. We further use our generated adversarial examples to improve models through adversarial training, and we demonstrate that our generated attacks are more robust against model re-training and different model architectures., Comment: 6 pages, accepted to EMNLP 2020

Image dehazing is a fast-growing research area in image processing and computer vision. Due to the extreme fog, haze, and air dispersion within an environment, the hazy image raises several challenges in retrieving original image information type. However, past techniques endure massive computation complexity and even the distortion of original images such as halos and over-saturation. In this research work, a new wavelet Hybrid (Local-Global Combined) Network is proposed for single image dehazing using a convolution neural network (CNN) in the wavelet domain (WH-Net). It is observed that low-level features such as edges are more important than high-level features such as texture. So we have used 2-DWT to decompose the single image model into the frequency subbands, which performs more quickly. It is demonstrated that the estimation of wavelet sub-bands reformulates the trainable end-to-end learning with a special architecture where DWT and IDWT are the feature extraction layers instead of Conv and Deconv, distinguishing it from classical CNN networks. The WHNet method is designed to achieve multi-level representations of hazy images to provide local and global information. The proposed network prominent features are designed with fewer convolution layers without decreasing performance relative to the commonly observed deeper learning models. Compared to several state-of-the-art algorithms, our proposed WHNet neural network surpasses in visual and quantitative performances on three public datasets.

Airborne LiDAR technology is a popular technique to quickly acquire high-precision information of the ground and the objects above it. Moreover, filtering the data is one of the necessary core processing steps. In recent years, many algorithms have been derived from traditional filtering algorithms. However, when applied to LiDAR data obtained from mountainous regions, most algorithms generate numerous omissions and errors when attempting to retain steep terrain features and filter vegetation information. This paper aims to quickly and accurately extract the digital terrain model (DTM) in mountainous regions. To accomplish this goal, a filtering algorithm based on a regular mesh generation strategy is proposed. The proposed algorithm initially divides the huge amount of point cloud data into strips and selects the appropriate spacing to subdivide each strip into equidistant grid data. The grid data is used as the input to an iterative polynomial fitting process, after which the point cloud is classified based on a controlled threshold. The experimental results show that the proposed method can quickly and efficiently classify data of mountains with different characteristics while retaining terrain feature information better than other algorithms. The average accuracy of recognition is greater than 92%. In addition, this algorithm also applies to mountains with lush vegetation.

Semantic segmentation in high-resolution remote-sensing (RS) images is a fundamental task for RS-based urban understanding and planning. However, various types of artificial objects in urban areas make this task quite challenging. Recently, the use of Deep Convolutional Neural Networks (DCNNs) with multiscale information fusion has demonstrated great potential in enhancing performance. Technically, however, existing fusions are usually implemented by summing or concatenating feature maps in a straightforward way. Seldom do works consider the spatial importance for global-to-local context-information aggregation. This paper proposes a Learnable-Gated CNN (L-GCNN) to address this issue. Methodologically, the Taylor expression of the information-entropy function is first parameterized to design the gate function, which is employed to generate pixelwise weights for coarse-to-fine refinement in the L-GCNN. Accordingly, a Parameterized Gate Module (PGM) was designed to achieve this goal. Then, the single PGM and its densely connected extension were embedded into different levels of the encoder in the L-GCNN to help identify the discriminative feature maps at different scales. With the above designs, the L-GCNN is finally organized as a self-cascaded end-to-end architecture that is able to sequentially aggregate context information for fine segmentation. The proposed model was evaluated on two public challenging benchmarks, the ISPRS 2Dsemantic segmentation challenge Potsdam dataset and the Massachusetts building dataset. The experiment results demonstrate that the proposed method exhibited significant improvement compared with several related segmentation networks, including the FCN, SegNet, RefineNet, PSPNet, DeepLab and GSN.For example, on the Potsdam dataset, our method achieved a 93.65% F 1 score and 88.06% I o U score for the segmentation of tiny cars in high-resolution RS images. As a conclusion, the proposed model showed potential for object segmentation from the RS images of buildings, impervious surfaces, low vegetation, trees and cars in urban settings, which largely varies in size and have confusing appearances.

How can we build recommender systems to take into account fairness? Real-world recommender systems are often composed of multiple models, built by multiple teams. However, most research on fairness focuses on improving fairness in a single model. Further, recent research on classification fairness has shown that combining multiple "fair" classifiers can still result in an "unfair" classification system. This presents a significant challenge: how do we understand and improve fairness in recommender systems composed of multiple components? In this paper, we study the compositionality of recommender fairness. We consider two recently proposed fairness ranking metrics: equality of exposure and pairwise ranking accuracy. While we show that fairness in recommendation is not guaranteed to compose, we provide theory for a set of conditions under which fairness of individual models does compose. We then present an analytical framework for both understanding whether a real system's signals can achieve compositional fairness, and improving which component would have the greatest impact on the fairness of the overall system. In addition to the theoretical results, we find on multiple datasets -- including a large-scale real-world recommender system -- that the overall system's end-to-end fairness is largely achievable by improving fairness in individual components., Comment: WSDM 2021

A multi-region scene matching-based localisation system for automated navigation of Unmanned Aerial Vehicles (UAV) is proposed. This system may serve as a backup navigation error correction system to support autonomous navigation in the absence of a global positioning system such as a Global Navigation Satellite System. Conceptually, the system computes the location of the UAV by comparing the sensed images taken by an on board optical camera with a library of pre-recorded geo-referenced images. Several challenging issues in building such a system are addressed, including the colour variability problem and elimination of time-varying details from the pairs of images. The overall algorithm is an iterative process involving four sub-processes: firstly, exact histogram matching is applied to sensed images to overcome the colour variability issues; secondly, regions are automatically extracted from the sensed image where landmarks are detected via their colour histograms; thirdly, these regions are matched against the library, while eliminating inconsistent regions between underlying image pairs in the registration process; and finally the location of the UAV is computed using an optimisation procedure which minimises the localisation error using affine transformations. Experimental results demonstrate the proposed system in terms of accuracy, robustness and computational efficiency.

Hydrometeor classification methods using polarimetric radar variables rely on probability density functions (PDFs) or membership functions derived empirically or by using electromagnetic scattering calculations. This paper describes an objective approach based on cluster analysis to deriving the PDFs. An iterative procedure with K-means clustering and expectation–maximization clustering based on Gaussian mixture models is developed to generate a series of prototypes for each hydrometeor type from several radar scans. The prototypes are then grouped together to produce a PDF for each hydrometeor type, which is modeled as a Gaussian mixture. The cluster-based method is applied to polarimetric radar data collected with the CP-2 S-band radar near Brisbane, Queensland, Australia. The results are illustrated and compared with theoretical classification boundaries in the literature. Some notable differences are found. Automated hydrometeor classification algorithms can be built using the PDFs of polarimetric variables associated with each hydrometeor type presented in this paper.

Current airborne laser scanning system is able to acquire three dimensional geographic information of areas and objects on the ground quickly with the form of three dimensional discrete point clouds which densities is up to 40 points per square meter. The main problem is the classification and identification of these objects in the point cloud. In this paper, we use the curved surface approximation method to classify point clouds into terrain and off-terrain points. A dimensionality fuzzy clustering method is introduced to classify buildings and vegetations. The method of surface approximation makes a distinguishing surface from the scanned data themselves, and makes the surface approach the ground by iteration. The surface made by his method fits the shape of the ground better. This paper compares the distinguishing effects of the method with and without dimensionality reduction, and result showed that the method with the dimensionality reduction is effective.

Inverse synthetic aperture radar (ISAR) imaging is conventionally addressed by means of fast Fourier transform (FFT) method. In this paper, a frame processing method derived based on frame theory for ISAR imaging is proposed as an alternative to FFT based approaches. In this method, a finite set of K consecutive target returns after a front-end processing by radar receiver are treated as a vector in a Hilbert space where a frame structure is defined. A frame and its dual are constructed as the functions of waveform and processing parameters and the target rotation rate. The latter can be estimated online. For certain classes of high resolution radar (HRR) waveforms, we show that it is possible to use standard methods of frame analysis and synthesis to reconstruct the ISAR image from radar returns. This method is analogous to those of reflective tomography, and does not suffer from the “range walk” problem exhibited by the FFT based approaches which poses constraints on the feasible aperture size for observing a target. Both simulated and experimental results are presented to demonstrate the effectiveness of the proposed method. Practical implementation and issues of the proposed frame processing technique are also discussed.

In target detection and tracking, the resolvability of multiple closely spaced targets of the sensor measurement is an important criterion for measuring the sensing ability of sensor systems like radar and sonar. The concept of information resolution for a sensor measurement system, which is defined in the framework of information geometry, is introduced. In particular, the information resolution of radar/sonar systems is generalised from the work on existing radar resolution pioneered by Woodward, et al. [1, 2] and defined on statistical manifolds where the intrinsic geometrical structure of both waveform, measurement, and noise models of the underlying sensing devices are conveniently characterized in terms of the Fisher information metric. This work is motivated by the fact that existing radar resolution identifies the ambiguities of the underlying waveforms regardless of the influence of noise but the latter often has significant contributions to outcomes such as target detection and tracking, radar imaging, etc. To this end, information resolution provides a unified statistical measure for the capability of sensing devices for a given application. Examples of basic radar measurements are presented to highlight the importance of information resolution for measurement systems in practical applications. Refereed/Peer-reviewed

Optimal determination of a UAV using a vision-based system to match images against a database is an important problem. It can be reformulated to the problem of using multiregion scene registration to match areas of a noisy and distorted image to a geo-referenced image. Under the assumptions that the mapping between sensed and geo-referenced images preserves gradients of straight lines cross mapping points on images and registration errors are all Gaussian distributed, we derive a two-stage weighted linear least square algorithm which localises the UAV optimally. Performance of the proposed algorithm is demonstrated via Monte Carlo multiple runs along with those available in literature.

Automatic optical and infrared (IR) image registration is a crucial requirement in the estimation of plant canopy temperature. The latter is used to compute the plant water stress status which can potentially be applied to an automated control system for optimal crop irrigation management. Due to the nature of image sources and plant scene, it is difficult to find enough common features between the pair of images to register them using a simple method. In this paper an improved registration algorithm is described, where a modified Sobel edge detector is adopted and a variable resolution scheme is applied. The algorithm can improve the registration efficiency considerably compared to our previous developed registration method. Experiment results show that the modified algorithm can achieve the same registration accuracy and a better success rate compared to that from the original algorithm with a substantial reduction of computational complexity.

Automatic registration of optical and IR images is a crucial step towards constructing an automated irrigation control system where plant water information is sensed via thermal imaging. The scene of the IR image is assumed to be completely included in the optical image and the alignment between the common scene in the two images may involve translation and rotation by a small angle, though a small scale difference may also be present. This automatic registration of data from two quite different, non-rigid imaging regimes presents several challenges, which cannot be overcome using common image processing techniques. In this paper, a fully automatic image registration algorithm for the alignment of optical and IR image pairs is described, where Pearson's cross-correlation between a pair of images serves as the similarity measure. A computationally efficient algorithm is designed and packaged as a software application. This work provides an intervention free process for extracting plant water stress information which can be fed into an automated irrigation scheduling program. The proposed algorithm is justified by the comparison of its registration performance with that of other potential algorithm techniques using several experimental data collections. Our results demonstrate the effectiveness of the proposed algorithm and efficiency of its application to the registration of IR and optical images.

In many radar and sonar tracking systems, the target state typically includes target position and velocity components that are estimated from a time sequence of target position and Doppler measurements. The use of measured Doppler information directly in the target trajectory estimation leads to a nonlinear filter implementation such as the extended Kalman filter (EKF), particle filter etc. We investigate a method for including Doppler measurements as part of the data association process and then assess the benefits of this approach. It is well understood that data association performance can dominate the total performance of a tracker that is designed to track targets in the presence of clutter. In this case, the Doppler component of the measurements may be used in combination with the target position measurement as an additional discriminant of measurement origin. We have developed a simple but efficient Doppler data association (DDA) method which utilises both position and Doppler measurements for single and multi-target tracking. If the Doppler measurements are not used in trajectory state estimation, then the nonlinear filters for the incorporation of Doppler measurements are not required, however a significant improvement in tracking performance is still observed. The proposed DDA method is demonstrated using both the linear multi-target integrated probabilistic data association algorithm (LMIPDA) and the linear multi-target integrated track splitting algorithm (LMITS) in an active sonar underwater multi-target tracking scenario.

We describe here a method for the analysis of materials on a conveyor belt using the natural gamma spectra collected with a BGO (Bismuth Germanate) gamma ray detector. This detector collects gamma ray emissions from the Potassium (K), Uranium (U), and Thorium (Th) atoms in the materials. Based on these data, and using a Poisson model for the data generation, a statistical model is proposed and an approximate maximum likelihood (ML) technique based on the expectation-maximization (EM) algorithm is then used to estimate the amount of each of the three elements in the material. The statistical model is further refined to incorporate parameters of drift in the detector and an estimation technique for this is developed and tested against real data. The Cramer-Rao lower bounds for the estimators are calculated.

When detecting a target using radar in the presence of multipath fading, the signal to noise ratio (SNR) at receiver can be dramatically reduced at certain ranges due to multipath signal cancellation, which leads to a significant drop of the detection probability at those ranges. The detection of a low elevation sea-surface target (LESST) is one example. For target tracking, this special case contrasts to the usual situation where we may often assume a constant probability of detection. Automatic target tracking requires a measure of track quality to discriminate between true and false tracks. This track quality measure should take both the existence of target and the SNR of the sensor into account, and be consistent with tracking performance. This paper investigates automatic target tracking using the integrated probabilistic data association (IPDA) type filters in the presence of multipath fading. Several existing single-scan probabilistic data association (PDA)-based filters, which calculate the probabilistic measure of track quality while estimating the track state, are evaluated in the scenario of tracking a nonmaneuvering LESST in the presence of both multipath fading and clutter. In addition, a variable Markov chain (VMC) IPDA tracker (IPDA-VMC), which has joint merits of both IPDA and the IPDA with a three state Markov chain (IPDA-M2), has been designed and evaluated here. In the derivation of track quality measure of both IPDA-M2 and 1PDA-VMC, the possibility that a target is present but is temporarily unable to be detected has been explicitly considered. A tracker using this track quality measure for false track discrimination has significantly improved LESST tracking performance. A simulation study verifies that both IPDA-M2 and IPDA-VMC have better capability to handle the LESST tracks and have an improved tracking performance over other PDA-based tracking algorithms under consideration.

Detection of weak signals in a low SNR environment is generally difficult, particularly, when the underlying signal noise is not only not Gaussian distributed but essentially unknown. A good example of such a case is the detection of termite biting signals from noisy audio data recorded by a passive acoustic sensor. In this paper, we present a novel technique to discriminate weak signals in data from noise of a learned non-Gaussian distribution. The proposed method, proceeds via the framework of generalised likelihood ratio test, and consists of two fundamental steps. First, an entropy-based incremental variational Bayesian inference is adopted to learn the non-Gaussian distribution from data using a Gaussian mixture model. An information geometric mapping of the data is then carried out via the total Bregman divergence (tBD), where the ambient noise distribution is approximated by the tBD-based l 1 -norm center of the neighboring data points over a specified time window. Experiment results show that the proposed method yields a significantly improvement in detection probability in low SNR and a robust detection performance compared with existing detection techniques.

Gating techniques for maneuvering target tracking using an IMM-PDA filter are considered in this paper. Existing gating techniques, namely, centralized gating and model based gating are reviewed and new gating techniques, called model probability weighted gating and two-stage model probability weighted gating, are proposed. A benchmark trajectory and a random scenario are considered for performance evaluation of the gating techniques in terms of RMS errors, percentage of track loss and computational load.

Knowledge of the target rotation rate is a crucial requirement for recovering the cross-range scale of a target image and compensating the image defocusing problem in ISAR imaging. This paper proposes a novel method for estimating the rotation rate of a target from the radar returns based on the frame processing imaging approach. The latter is an alternative ISAR imaging technique to the fast Fourier transform based approaches and target rotation rate is explicitly embedded in the frame processing structure. The method utilizes the intrinsic structures of the frame which spans the Hilbert space of radar return from which the rotational information of target is extracted. (5 pages)

Frame processing method offers a model-based approach to Inverse Synthetic Aperture Radar (ISAR) imaging. It also provides a way to estimate the rotation rate of a non-cooperative target from radar returns via the frame operator properties. In this paper, the relationship between the best achievable ISAR image and the reconstructed image from radar returns was derived in the framework of Finite Frame Processing theory. We show that image defocusing caused by the use of an incorrect target rotation rate is interpreted under the FP method as a frame operator mismatch problem which causes energy dispersion. The unknown target rotation rate may be computed by optimizing the frame operator via a prominent point. Consequently, a prominent intensity maximization method in FP framework was proposed to estimate the underlying target rotation rate from radar returns. In addition, an image filtering technique was implemented to assist searching for a prominent point in practice. The proposed method is justified via a simulation analysis on the performance of FP imaging versus target rotation rate error. Effectiveness of the proposed method is also confirmed from real ISAR data experiments.

When tracking a target using phase-only signal returns, range ambiguities are a major issue. In this work, a look-up table between phase measurement space and target location space is constructed for phase measurement mapping. Solving such problems via solutions to Diophantine equations has been used to locate candidate locations. Here we show how such target location ambiguity can also be resolved over time when the underlying target is in motion and where issues of clutter are treated via a phase distribution discrimination method. That is, a probability density function of the ambiguous phase-only measurement that takes both sensor noise and target motion distributions into account is derived based on directional statistics. This approach to solving phase ambiguity under significant clutter conditions has promising results. Refereed/Peer-reviewed

When tracking a target using phase-only signal returns, range ambiguities due to the modulo 2π in measured signal phases are a major issue. Standard approaches to such problems would typically involve the use of Diophantine equations. In this paper, a simpler and robust solution is examined which uses look-up tables defined between the phase measurement and target location spaces to determine the phase measurement mapping. We show, first, that when the target motion is significant between data sampling intervals the location ambiguity can be resolved over time via known target-in-cluster tracking techniques. This method determines the optimal allocation of location with respect to phase measurements relative to the quantization of look up table values. Second, when the target is undergoing micromotions (jitter) which results in the same collection of candidate locations from phase measurements over time, the location ambiguity can be resolved using a novel phase distribution discrimination method. In this method a probability density function of the ambiguous phase-only measurement is derived that takes both sensor noise and target motion distributions into account based on directional statistics. Optimal locations are inferred from such distributions. Examples are given to demonstrate the effectiveness of these proposed methods. Refereed/Peer-reviewed

Object tracking in video surveillance is often restrained by the requirement of real time processing. In this paper, the object tracking problem is approached as two independent processes. At each video frame, multi-object information is approximately represented by a set of rectangle patches as the result of object detection and the data set is converted to a set of virtual location measurements as well as the associated feature measurements with assumption that location and feature measurements are independent. In the consequent process, a feature aided integrated probabilistic data association type multi-object tracker is used to recursively update the posterior densities of the objects estimated frame by frame based on the virtual measurements obtained from the former process. The uncertainty of the converted virtual measurements largely depends on the object detection techniques used and it may be handled in the later process by the tracker. Our results demonstrated that the feature aided data association technique can resolve the uncertainty due to the imperfect of virtual measurement process. Furthermore, the required computational overhead is considerably less than those of pixel-wise based approaches.

PMHT algorithm, as proposed, promises high performance multi target tracking in clutter with (relatively) modest computational resources. However, when applied to practical target tracking situations, a number of problems need to be overcome. PMHT assumes fixed number of tracks, and furthermore it assumes that all tracks are true tracks. No track quality measure is provided within PMHT to enable false track discrimination. In this paper we add the probability of target existence to PMHT processing, and accommodate a variable number of tracks.

In many radar or sonar tracking systems, where the state of interest typically includes target position and velocity components, target Doppler measurements may be available in addition to the target position measurements. Using measurements of Doppler information directly in target trajectory estimation leads to a nonlinear filter implementation. In this paper we investigate the methods of using Doppler measurements to improve tracking performance while maintaining the tracker implementation in the linear domain. Two new methods, namely, the one point track initialisation and the Doppler data association are proposed. By applying these two methods on the linear multi-target integrated probabilistic data association algorithm (LMIPDA), we improve tracking performance in terms of reduced number of false tracks and computational load. The merits of the proposed approaches are demonstrated via an active sonar underwater multi-target tracking scenario.

Energy based detection measures sensor received signal strength (RSS) transmitted from a target. In this paper, we propose a new approach for estimating a moving target trajectory over a sensor field via energy based detections as an alternative to trilateration positioning or nonlinear estimation. In 2D case, possible target locations described by a RSS ratio from two sensors are approximated using a set of Gaussian random variables which are refereed to as location measurements. At each data collection time, several sets of such measurements can be found from RSS ratios which are due to multiple sensor detections. A track splitting filter is used to perform either measurement fusion and target state estimation using these measurements. The RSS ratio data mapping via Gaussian density approximation plays a key role in the proposed target tracking method and is robust in the sense that it can tolerate larger RSS noise and using additional sensor detections to improve tracking performance over trilateration based techniques. The effectiveness of the propose method is demonstrated via an example of tracking a moving target over a sensor network of small acoustic sensors.

In this paper, a one-point track initiation method is derived for conventional target tracking systems where noisy sensor measurements of both target position and Doppler are available. Tentative tracks are initiated using position measurements from every single scan and the associated Doppler measurements are incorporated to reduce the velocity uncertainties of these initiated track states. The proposed track initiation method is capable of reducing the computational requirements of the system quadratically compared to the classical two-point method when false measurements are present. We further demonstrate that a tracking system which incorporates Doppler measurements both for track initiation and measurement-to-track association can greatly reduce the number of false tracks that are generated and provides greater flexibility for confirming true tracks without sacrificing the system's performance in terms of the number of confirmed false tracks. As a consequence, a tracking system that incorporates the proposed method will exhibit a much shorter true track confirmation delay than a similar system based on the standard approach.

The active sonar environment presents a challenge for any tracking system. Multipath propagation and the fact that targets are, in general, extended objects mean that multiple measurements may be obtained for each target in the surveillance region. In addition, other objects and random scatterers introduce clutter measurements. Whilst pre-processing such as clustering and classification of detections can assist in significantly reducing the number of measurements, there may still be multiple measurements resulting from a single target. This also applies in a multitarget environment, where multiple measurements may be received from each target. Automatic target tracking, which uses measurements in consecutive scans to initiate tracks, generate a large number of false tracks. The key problem for a tracker becomes that of false track discrimination, i.e. confirming true tracks and terminating false tracks. This paper investigates the performance of three trackers for multitarget tracking in the active sonar environment, each using a different data association paradigm. These are the global nearest neighbor (GNN) data association, linear multitarget integrated probabilistic data association (LMIPDA) and nearest neighbour joint probabilistic data association (NNJPDA). A simulation study of these algorithms in a simulated active sonar environment is presented.

Making use of information acquired from a sensor network to improve the accuracy of target tracking is one of the most important issues in sensor network research. This paper demonstrates this philosophy using a distributed dissimilar sensor fusion scenario, where a tracker was established and maintained by a surface radar sensor and the distributed sensor fusion is performed whenever target measurement from an angle-only sensor is available to the radar sensor. The target state information is extracted from the angle-only sensor measurement so that the distributed track fusion at radar sensor can be performed. The extended Kalman filters (EKF) have been used to implement all tracking functions due to the nonlinearity between target state and the associated sensor observations. The scenario is conveniently implemented using advanced radar tracking system (ARTS) toolbox in Matlab Simulink environment. Our simulation results have shown the improvement of the (racking accuracy bv applying distributed track fusion. The convenience of using ARTS toolbox for complex algorithm implementation and testing are also clear from the context.

A method is proposed for estimating the target rotation rate for ISAR imaging based on the frame processing technique. The method utilises the intrinsic structures of the frame which span the Hilbert space of radar return from which the rotational information of the target is extracted. For a prominent point, the intensity of the projection is maximum when the frame component corresponding to the prominent point is built with the target rotation rate.

The out-of-sequence measurement (OOSM) problem is one of the most important issues for target tracking in a multisensor fusion network. The optimal solution is to augment the state vector to include the past states which correspond to those delayed measurements in the filtering process. In this paper, maneuvering target tracking in clutter with the OOSM problem is considered. An AS-IMM-PDA algorithm that uses OOSM to improve tracking performance is presented. The benefit of the AS-IMM-PDA is demonstrated via a maneuvering target tracking example.

Information geometry, as a powerful though complex mathematical tool can provide additional insights in the analysis of sensor measurement. In this paper, the application of information geometry to the analysis of sensor networks is explored in an attempt to gain a better understanding of sensor system issues for target detection and tracking. In particular, the (integrated) Fisher information distance between two targets is used to measure target resolvabil-ity in the region covered by the sensor system and is approximately calculated. It is also compared with the Kullback Leibler divergence. The proposed analysis is elucidated via two simple sensor network examples in the context of target tracking. The preliminary analysis results presented in this paper provide evidence that information geometry is able to offer consistent but more comprehensive means to understand and solve sensor network problems which are difficult to deal with via conventional analysis methods.