Target Classification Information Research Articles

Clustering ensemble-based novelty score for outlier detection

Recently, One-class classification algorithms have been successfully used for outlier detection problems in several industrial fields. However, in case of that the target class has complex structures, single outlier detection model with one-class classifier often poorly performs because it cannot appropriately reflect intrinsic data structures. To address this limitation, we propose a clustering ensemble-based novelty score algorithm. The proposed algorithm calculates novelty score from the mixture of multiple clustering solutions generated by both random subspace and random-K ensemble approaches. Then, final ensemble novelty score is defined by summarizing multiple novelty scores obtained from individual clustering results. Because these multiple novelty scores are computed from many possible characteristics of target class information, the proposed ensemble novelty score can appropriately reflect the inherent structures of target class. Experiments were conducted on various benchmark datasets to compared with existing methods and investigate the properties of the proposed algorithm. The experimental results confirm that the proposed algorithm outperforms existing one-class classification methods in various cases.

Optimization Algorithm of Moving Object Detection Using Multiscale Pyramid Convolutional Neural Networks.

Object detection and recognition is a very important topic with significant research value. This research develops an optimised model of moving target identification based on CNN to address the issues of insufficient positioning information and low target detection accuracy (convolutional neural network). In this article, the target classification information and semantic location information are obtained through the fusion of the target detection model and the depth semantic segmentation model. The classification and position portion of the target detection model is provided by the simultaneous fusion of the image features carrying various information and a pyramid structure of multiscale image features so that the matched image fusion characteristics can be used by the target detection model to detect targets of various sizes and shapes. According to experimental findings, this method's accuracy rate is 0.941, which is 0.189 higher than that of the LSTM-NMS algorithm. Through the migration of CNN and the learning of context information, this technique has great robustness and enhances the scene adaptability of feature extraction as well as the accuracy of moving target position detection.

Real Micro-Doppler Parameters Extraction of Spinning Targets Based on Rotating Interference Antenna

Micro-Doppler is a unique characteristic of targets with micro-motions, which can provide significant information for target classification and recognition. However, the monostatic radar has the shortcoming of only obtaining the radial micro-motion characteristics. Although the vortex-electromagnetic-wave-based radar has the potential to obtain real micro-motion parameters, it has a high dependence on the mode number and purity of the orbital angular momentum, which greatly restricts its application in the micro-motion parameter extraction. To overcome the above problems, a new radar configuration based on the rotating interference antenna is proposed in this paper. Through the interference processing of the micro-Doppler curves of the rotating and fixed antenna, the curves containing the real micro-motion information of the target can be obtained. Then the real micro-motion characteristics of the spinning target can be reconstructed by the orthogonal matching pursuit algorithm. The effectiveness and robustness of the proposed method are validated by simulations.

Multimodal object tracking by exploiting appearance and class information

In this work, we study the method exploiting natural language network to improve tracking performance. We propose a novel architecture which can combine class and visual information presented in tracking. To this end, we introduce a multimodal feature association network, allowing us to correlate the target class with its appearance during training and aid the localization of the target during inference. Specifically, we first utilize an appearance model to extract the target visual features, from which we obtain appearance cues, for instance shape and color. In order to employ target class information, we design a learned lightweight embedding network to embed the target class into a feature representation. The association network of our architecture contains a multimodal fusion module and a predictor module. The fusion module is used to combine features from class and appearance, yielding multimodal features with more expressive representations for the subsequent module. The predictor module is used to determine the target location in the current frame, from which we associate the class to the appearance. The class embedding module thus can learn appearance cues by exploiting the back-propagation functionality. To verify the abilities of our method, we select the official training and test splits of the LaSOT with annotated images and classes to perform experiments. In particular, we analyze the imbalance in the samples and employ a class validator discriminator to alleviate this problem. Extensive experimental results on LaSOT, UAV20L and UAV123@10fps demonstrate our method achieves competitive results while maintaining a considerable real-time speed.

An Adversarial Framework for Generating Unseen Images by Activation Maximization

Activation maximization (AM) refers to the task of generating input examples that maximize the activation of a target class of a classifier, which can be used for class-conditional image generation and model interpretation. A popular class of AM method, GAN-based AM, introduces a GAN pre-trained on a large image set, and performs AM over its input random seed or style embeddings, so that the generated images are natural and adversarial attacks are prevented. Most of these methods would require the image set to contain some images of the target class to be visualized. Otherwise they tend to generate other seen class images that most maximizes the target class activation. In this paper, we aim to tackle the case where information about the target class is completely removed from the image set. This would ensure that the generated images truly reflect the target class information residing in the classifier, not the target class information in the image set, which contributes to a more faithful interpretation technique. To this end, we propose PROBEGAN, a GAN-based AM algorithm capable of generating image classes unseen in the image set. Rather than using a pre-trained GAN, PROBEGAN trains a new GAN with AM explicitly included in its training objective. PROBEGAN consists of a class-conditional generator, a seen-class discriminator, and an all-class unconditional discriminator. It can be shown that such a framework can generate images with the features of the unseen target class, while retaining the naturalness as depicted in the image set. Experiments have shown that PROBEGAN can generate unseen-class images with much higher quality than the baselines. We also explore using PROBEGAN as a model interpretation tool. Our code is at https://github.com/csmiler/ProbeGAN/.

Aerial target classification using an LFL-Net with multiview HRRPs

ABSTRACT Non-cooperative aerial target classification is one of the most attractive but challenging tasks in radar remote sensing applications. Multiview high-range resolution profiles (HRRPs) of the aerial target contain abundant information and will benefit to classification. In this paper, a new aerial target classification method based on an end-to-end lightweight feature learning network (LFL-Net) with multiview HRRPs is proposed. The aerial target classification scenario using multiview HRRPs is first studied and modelled. Then a LFL-Net with multi-inputs and some distinct modules is designed to effectively learn the target classification information from the multiview HRRPs. Therefore, the proposed method can achieve accurate and reliable classification results under different signal-to-noise ratios (SNRs). Experimental results have shown the superiorities of the proposed non-cooperative aerial target classification method.

Tracking of Evasive Objects Using Bistatic Doppler Radar Operating in the Millimeter Wave Regime

In this study, we propose a range detection (RD) ability by a continuous wave (CW) bistatic Doppler radar (RDCWB) of small and fast targets with very high range resolution. The target’s range and velocity are detected simultaneously. The scheme is based on the transmission of a continuous wave (CW) at millimeter wavelength (MMW) and the measurement of the respective Doppler shifts associated with target movements in different directions. The range resolution in this method is determined by the Doppler resolution only, without the necessity to transmit the modulated waveforms as in frequency modulation continuous wave (FMCW) or pulse radars. As the Doppler resolution in CW depends only on the time window required for processing, a very highrange resolution can be obtained. Most other systems that perform target localization use the transmission of wide-band waveforms while measuring the delay of the received signal scattered from the target. In the proposed scheme, the range resolution depends on the processed integration time of the detected signal and the velocity of the target. The transmission is performed from separated antennas and received by a single antenna. The received signal is heterodyned with a sample of the transmitted signal in order to obtain the Doppler shifts associated with the target’s movement. As in a multi-in multi-out (MIMO) configuration, the presented scheme allows for the accumulation of additional information for target classification. Data on the target’s velocity, distance, direction, and instantaneous velocity can be extracted. Using digital processing, with the additional information obtained by analyzing the difference between the resulting intermediate frequencies caused by the Doppler effect, it is possible to calculate the distance between the radar and the target at high resolution in real-time. The presented method, which was tested experimentally, proved to be highly effective, as only one receiver is required for the detection, while the transmission is carried out using a fixed, single-frequency transmission.

Rotational Angular Velocity Estimation of Rotor Target via 3D-OMP-Based Parametric Sparse Representation

As a special feature of rotor target, rotational angular velocity can provide important information for target classification and recognition. In this paper, a rotational angular velocity estimation method is proposed based on complex empirical mode decomposition (CEMD) and three-dimensional orthogonal matching pursuit (3D-OMP)-based parametric sparse representation (PSR) technique. Firstly, the translational radial velocity is estimated from radar echo via Fourier transform, and then the translational compensation is implemented according to the estimated velocity. Secondly, in order to eliminate the influence of the target main body echo on micro-Doppler parameter estimation, CEMD algorithm is adopted for target echo separation. Then, the separated signal of rotor echo is formulated as a jointly sparse signal through a three-dimensional parametric dictionary matrix, which converts the rotational angular velocity estimation into a problem of dynamic representation of jointly sparse signals. Finally, in order to reduce the computational complexity of the estimation method, we propose a 3D-OMP-based parametric sparse representation algorithm to achieve the sparse solution, and the rotational angular velocity can be estimated by minimizing the reconstruction error. The experimental results verify the effectiveness of the proposed algorithm.

Target Classification Aided Variable-Structure Multiple-Model Algorithm

Because the varieties of general aviation aircraft and their performance differences become increasingly large, the traditional multiple model tracking algorithm needs to employ more motion models to describe the actual maneuver model of a moving target. This fact is easy to degrade the tracking accuracy and brings a large computational load. Thus, a target classification aided variable-structure multiple-model algorithm (TCA-VSMM) is proposed to solve the above problem. In the proposed TCA-VSMM algorithm, the target classification aided is introduced to improve the accuracy of state estimation. Then, the target classification aided in the screening of the motion model set is analyzed. Concretely, the target classification information and velocity information in automatic dependent surveillance-broadcast (ADS-B) measurements are incorporated into the variable-structure multiple-model filter. Consequently, the screening model set is more approximated to the real motion model of a moving target. Experiments show that the TCA-VSMM algorithm can obtain better performance with small computation load and high estimation accuracy than those of the model-group switching variable-structure multiple-model algorithm.

POSE.3C: Prediction-Based Opportunistic Sensing Using Distributed Classification, Clustering, and Control in Heterogeneous Sensor Networks

This paper presents a distributed algorithm, called prediction-based opportunistic sensing using distributed classification, clustering, and control (POSE.3C), for self adaptation of sensor networks for energy management. The underlying 3C network autonomy concept enables utilization of the target classification information to form dynamic clusters around the predicted target position via selection of sensor nodes with the highest energies and maximum geometric diversity. Furthermore, the nodes can probabilistically control their heterogeneous devices to track targets of interest and minimize energy consumption in a completely distributed manner. Theoretical properties of the POSE.3C network are established and derived in terms of the network lifetime and missed detection characteristics. The algorithm is validated through extensive simulations that demonstrate a significant increase in the network lifetime as compared to other network control approaches, while providing high tracking accuracy and low missed detection rates.

Optimized Algorithm for Target Tracking using Classification Information based Association Filter

In last few decades, multiple target tracking fetches quite attention to the researchers for object localization and monitoring target trajectories which has become one of the most used technique in the area of visual tracking, traffic monitoring, air surveillance system, robotics and vision. On the basis of S-D assignment algorithm, a new algorithm for tracking multiple targets in presence of clutter is designed. By considering target classification information received as special feature from target scan report, cost coefficients of dynamic assignment matrix are modified accordingly using joint probabilistic data association filter. The tracking results get improved with the use of target class and kinematic features information where the association costs are similar for different targets. With the help of the information collected in current scan the classifier output is dynamically updated to incorporate new target classes to be used future scans. Simulation results show that new algorithm can attain competitive tracking performance with distributed computational load by utilizing target classification information into dynamic multidimensional assignment algorithm. The main contribution of this paper is the development of new target tracking method based on IMM filter which generate dynamic classifier to incorporate target features information. This additional information about targets present in current scan helps to take future scans data association decisions.

Long-range target detection and classification system for environmental monitoring at marine hydrokinetic (MHK) sites

BioSonics and team members are developing a practical, unobtrusive, robust, and cost-effective long range (200–300 m) active acoustic monitoring system to automatically assess marine life behavior at potential or operational MHK sites. The proposed system includes a Perimeter Detector, that will automatically detect and geolocate targets at ranges of 200–300 m. The system will include a Directed Classifier that will be automatically aimed at a detected target to track the target’s position in three dimensions. This tracking capability allows automated measurement of the target’s behavior (i.e., speed, direction, and tortuosity), a strong indicator of target classification. Acoustic signatures from tracked targets will be analyzed to provide additional target classification information. Low band width, real-time reports will be automatically generated and transmitted to project operators, including target location, depth, behavior, and classification.

Precession parameter estimation method of the cone‐shaped target based on inverse synthetic aperture radar image sequence analysis

The micro-motion features induced by space targets provide effective information for target classification and recognition. This study introduces the equivalent rotational velocity (ERV) estimation for precession parameter estimation of the cone-shaped target with inverse synthetic aperture radar image sequence. On the basis of the rotation matching relationship of precession cone-shaped target, the estimated ERV is used to build an eigenfunction, and the parameter searching set is obtained. Through theoretical analysis, the coupled parameters of the eigenfunction will concentrate on a curve on the searching plane. Thus, the parameter searching set is reduced to the neighbouring area of this curve, which reduces the computational burden greatly. Finally, the parameters are decoupled by the reconstructed radar observation angle, and the precession parameters such as precession period, precession angle, average observation angle, radial length and bottom radius of the target are precisely estimated. The proposed algorithm requires no prior information and it has high computational efficiency. Simulation results with electromagnetic data illustrate the validity of the proposed method.

Clustering-based proxy measure for optimizing one-class classifiers

One-class classification is a type of unsupervised learning task wherein only information on the target class is available and that on other classes is not. As this corresponds to many real-world data scenarios, considerable research efforts have been focused on developing one-class classification algorithms to build well-performing one-class classifiers. The performance of one-class classifiers largely depends on their hyperparameters. However, optimizing hyperparameters is inherently difficult and challenging because conventional validation procedures are not directly applicable in the absence of non-target class information. The existing methods have focused on specific one-class classification algorithms, and therefore, could not be extended to other algorithms. In this study, we present a clustering-based proxy measure to evaluate the quality of different hyperparameter candidates for a one-class classification algorithm when only the target class information is given. The proposed proxy measure utilizes conventional measures of classification performance by introducing pseudo non-target classes that are derived based on clustering. The best hyperparameters for the algorithm can be estimated by alternatively optimizing this measure. It does not require incorporating any prior knowledge, and can be implemented with various types of one-class classification algorithms. We demonstrate the effectiveness through experiments comprising benchmark problems.

Three-dimensional interferometric imaging and precession feature extraction of space targets in wideband radar

Three-dimensional (3-D) imaging and the 3-D micromotion feature extraction technique of space targets provide significant information for target classification and recognition. The idea of multiantenna interferometric processing from the interferometric inverse synthetic aperture radar (InISAR) is introduced into the 3-D imaging of micromotional space targets. After the modeling analysis of the micromotion characteristic induced by the target on the range–slow-time plane, a 3-D interferometric imaging method for space precession targets in wideband radar is proposed. The real 3-D images of the target in both the normal mode and the squint mode are obtained by the proposed method. Meanwhile, some 3-D precession features and structure parameters of the target are extracted accurately. Finally, simulation results validate the effectiveness and robustness of the proposed method. Compared with the existed approaches, this study can achieve real 3-D imaging and extract precession features and structure parameters for space precession target only by a monostatic radar with multi-antenna.

Compressive phase-only filtering at extreme compression rates

We introduce an efficient method for the reconstruction of the correlation between a compressively measured image and a phase-only filter. The proposed method is based on two properties of phase-only filtering: such filtering is a unitary circulant transform, and the correlation plane it produces is usually sparse. Thanks to these properties, phase-only filters are perfectly compatible with the framework of compressive sensing. Moreover, the lasso-based recovery algorithm is very fast when phase-only filtering is used as the compression matrix. The proposed method can be seen as a generalization of the correlation-based pattern recognition technique, which is hereby applied directly to non-adaptively acquired compressed data. At the time of measurement, any prior knowledge of the target object for which the data will be scanned is not required. We show that images measured at extremely high compression rates may still contain sufficient information for target classification and localization, even if the compression rate is high enough, that visual recognition of the target in the reconstructed image is no longer possible. The method has been applied by us to highly undersampled measurements obtained from a single-pixel camera, with sampling based on randomly chosen Walsh–Hadamard patterns.

AN ACTIVITY PREDICTION MODEL USING SHAPE-BASED DESCRIPTOR METHOD

Similarity searching, the activity of an unknown compound (target) is predicted through the comparison of an unknown compound with a set of known activities of compounds. The known activities of the most similar compounds are assigned to the unknown compound. Different machine learning methods and Multilevel Neighborhoods of Atoms (MNA) structure descriptors have been applied for the activities prediction. In this paper, we introduced a new activity prediction model with Shape-Based Descriptor Method (SBDM) .Experimental results show that SBDM-MNA provides a useful method of using the prior knowledge of target class information (active and inactive compounds) of predicting the activity of orphan compounds. To validate our method, we have applied the SBDM-MNA to different established data sets from literature and compare its performance with the classical MNA descriptor for activity prediction.

Joint detection, tracking and classification of multiple maneuvering targets based on the linear Gaussian jump Markov probability hypothesis density filter

This paper is to account for joint detection, tracking, and classification (JDTC) of an unknown and time-varying number of maneuvering targets in clutter. Unlike most tracking algorithms that use only the kinematic measurements, this paper proposes a recursive JDTC algorithm to exploit the coupled information between target detection, tracking and classification based on the Gaussian mixture probability hypothesis density filter (GMPHDF) in the linear Gaussian jump Markov systems (LGJMS) multitarget model. The original LGJMS-GMPHDF, devoted to joint detect and track all potential targets utilizing an identical and fixed set of models, has been modified to incorporate target class information and the class-dependent kinematic model set. The mutual dependence between target kinematics and class is exploited twice: first to construct the combined model sets, then to compute the combined measurement likelihood. The proposed algorithm is illustrated via a simulation example involving tracking of two closely spaced parallel moving targets and two crossing moving targets from different classes, where targets can appear and disappear.

基于MMPHDF 的多机动目标联合检测、跟踪与分类算法

To account for joint detection, tracking and classification (JDTC) of multiple maneuvering targets in dense clutter environment, this paper introduces an algorithm based on the multiple-model probability hypothesis density filter (MMPHDF). The MMPHDF can be applied to jointly detect and track multiple maneuvering targets, but its filter performance deteriorates rapidly in dense clutter environment. The proposed JDTC algorithm (MMPHDF-JDTC) utilizes target classification information and target kinematic information to simultaneously estimate the time-varying number of targets, their kinematic states and types. The main idea is to augment the kinematic state vector with the target class vector, and then use their combined measurement likelihood to integrate the target classification information into the update process of MMPHDF. The combined target kinematic state and class measurement likelihood improves the discrimination of different class targets and clutter, so better detection and tracking performance can be expected compared with the original MMPHDF. Alternately, accurate detection and tracking results is the foundation for correct target classification. A particle implementation of the MMPHDF-JDTC has been given. Simulation results validate the above conclusions.

음향 표적 식별을 위한 무선 센서 네트워크에서 웨이블릿 상수를 이용한 표적 특징 추출

Acoustic target classification in wireless sensor network is important research at environmental surveillance, invasion surveillance, multiple target separation. General sensor node signal processing methods concentrated on received signal energy based target detection and received raw signal compression. The former is not suited to target classification because of almost every target information are lost except target energy. The latter bring down life-time of sensor node owing to high computational complexity and transmission energy. In this paper, we introduce an feature extraction algorithm for acoustic target classification in wireless sensor network which has time and frequency information. The proposed method extracts time information and de-noised target classification information using wavelet decomposition step. This method reduces communication energy by 28% of original signal and computational complexity.