Search

The Adaptive Boosting (AdaBoost) algorithm is a widely used ensemble learning framework, and it can get good classification results on general datasets. However, it is challenging to apply the AdaBoost algorithm directly to imbalanced data since it is designed mainly for processing misclassified samples rather than samples of minority classes. To better process imbalanced data, this paper introduces the indicator Area Under Curve (AUC) which can reflect the comprehensive performance of the model, and proposes an improved AdaBoost algorithm based on AUC (AdaBoost-A) which improves the error calculation performance of the AdaBoost algorithm by comprehensively considering the effects of misclassification probability and AUC. To prevent redundant or useless weak classifiers the traditional AdaBoost algorithm generated from consuming too much system resources, this paper proposes an ensemble algorithm, PSOPD-AdaBoost-A, which can re-initialize parameters to avoid falling into local optimum, and optimize the coefficients of AdaBoost weak classifiers. Experiment results show that the proposed algorithm is effective for processing imbalanced data, especially the data with relatively high imbalances.

This paper presents a robust adaptive beamforming algorithm based on an attention convolutional neural network (ACNN) for coprime sensor arrays, named the CAWE-ACNN algorithm. In the proposed algorithm, via a spatial and channel attention unit, an ACNN model is constructed to enhance the features contributing to beamforming weight vector estimation and to improve the signal-to-interference-plus-noise ratio (SINR) performance, respectively. Then, an interference-plus-noise covariance matrix reconstruction algorithm is used to obtain an appropriate label for the proposed ACNN model. By the calculated label and the sample signals received from the coprime sensor arrays, the ACNN is well-trained and capable of accurately and efficiently outputting the beamforming weight vector. The simulation results verify that the proposed algorithm achieves excellent SINR performance and high computation efficiency.

Human being detection via ultra-wideband (UWB) radars has shown great prospects in many areas, such as biomedicine, military operation, public security, emergency rescue, and so on. When a person stays stationary, the main feature that separates him/her from surroundings is the movement of chest wall due to breath. There have been many algorithms developed for breath detection while using UWB radars. However, those algorithms were almost based on a basic scheme that focused on processing in the time dimension of UWB data. They did not utilize the benefits from the wide operational bandwidth of UWB radars to show potential superiority over those narrowband systems such as a continuous wave (CW) Doppler radar. In this paper, a breath detection method was proposed based on operational bandwidth segmentation. A basic theoretical model was firstly introduced, indicating that characteristics of breath signals contained in UWB echoes were consistent among the operational frequencies, while those of clutters were not. So, the method divided a set of UWB echo data into a number of subsets, each of which corresponded to a sub-band within the operational bandwidth of the UWB radar. Thus information about the operational frequency is provided for subsequent processing. With the aid of the information, a breath enhancement algorithm was developed mainly by averaging the segmented UWB data along the operational frequency. The algorithm’s performance was verified by data measured by a stepped-frequency CW (SFCW) UWB radar. The experimental results showed that the algorithm performed better than that without the segmentation. They also showed its feasibility for fast detection of breath based on a short duration of data. Moreover, the method’s potential for target identification and impulse-radio (IR) UWB radar was investigated. In summary, the method provides a new processing scheme for UWB radars when they are used for breath detection. With this scheme, the UWB radars have a benefit of greater flexibility in data processing over those narrowband radars, and thus will perform more effectively and efficiently in practical applications.

Ultra-wide band (UWB) radar for short-range human target detection is widely used to find and locate survivors in some rescue missions after a disaster. The results of the application of bistatic UWB radar for detecting multi-stationary human targets have shown that human targets close to the radar antennas are very often visible, while those farther from radar antennas are detected with less reliability. In this paper, on account of the significant difference of frequency content between the echo signal of the human target and that of noise in the shadowing region, an algorithm based on wavelet entropy is proposed to detect multiple targets. Our findings indicate that the entropy value of human targets was much lower than that of noise. Compared with the method of adaptive filtering and the energy spectrum, wavelet entropy can accurately detect the person farther from the radar antennas, and it can be employed as a useful tool in detecting multiple targets by bistatic UWB radar.

Remarkable progress has been achieved in the detection of single stationary human. However, restricted by the mutual interference of multiple humans (e.g., strong sidelobes of the torsos and the shadow effect), detection and localization of the multiple stationary humans remains a huge challenge. In this paper, ultra-wideband (UWB) multiple-input and multiple-output (MIMO) radar is exploited to improve the detection performance of multiple stationary humans for its multiple sight angles and high-resolution two-dimensional imaging capacity. A signal model of the vital sign considering both bi-static angles and attitude angle of the human body is firstly developed, and then a novel detection method is proposed to detect and localize multiple stationary humans. In this method, preprocessing is firstly implemented to improve the signal-to-noise ratio (SNR) of the vital signs, and then a vital-sign-enhanced imaging algorithm is presented to suppress the environmental clutters and mutual affection of multiple humans. Finally, an automatic detection algorithm including constant false alarm rate (CFAR), morphological filtering and clustering is implemented to improve the detection performance of weak human targets affected by heavy clutters and shadow effect. The simulation and experimental results show that the proposed method can get a high-quality image of multiple humans and we can use it to discriminate and localize multiple adjacent human targets behind brick walls.

The through-wall detection and classification of human activities are critical for anti-terrorism, security, and disaster rescue operations. An effective through-wall detection and classification technology is proposed for finer-grained human activities such as piaffe, picking up an object, waving, jumping, standing with random micro-shakes, and breathing while sitting. A stepped-frequency continuous wave (SFCW) bio-radar sensor is first used to conduct through-wall detection of finer-grained human activities; Then, a comprehensive range accumulation time-frequency transform (CRATFR) based on inverse weight coefficients is proposed, which aims to strengthen the micro-Doppler features of finer activity signals. Finally, in combination with the effective eigenvalues extracted from the CRATFR spectrum, an optimal self-adaption support vector machine (OS-SVM) based on prior human position information is introduced to classify different finer-grained activities. At a fixed position (3 m) behind a wall, the classification accuracies of six activities performed by eight individuals were 98.78% and 93.23%, respectively, for the two scenarios defined in this paper. In the position-changing experiment, an average classification accuracy of 86.67% was obtained for five finer-grained activities (excluding breathing) of eight individuals within 6 m behind the wall for the most practical scenario, a significant improvement over the 79% accuracy of the current method.