8 December 2023 in IEEE Transactions on Instrumentation and Measurement
In this article, a multisensor detection system is constructed for Herbal Medicine odor detection and recognition based on a weighted scheme with area under curve (AUC) and information theory. In the feature extraction stage, the multifeature combination that is more conducive to classification is selected by comparison to lay a good foundation for subsequent sensor array optimization. In the array optimization stage, the performance of the array optimization algorithm on unbalanced datasets is further improved by developing the joint mutual information (JMI) and AUC measure (AJMI) based on the consideration of correlation and redundancy among features. Experimental results demonstrate the superiority of the proposed design in comparison to other methods from the literature. In particular, it is verified that the proposed AJMI algorithm can improve recognition accuracy by 4% and $F1$ score by more than 4% while reducing the number of sensors from 37 to 23 compared with the multiple-feature extraction design without considering array optimization. It is worth noting that the proposed design achieves better recognition results than other algorithms when using only 14 features from ten sensors.
22 December 2025 in IEEE Transactions on Wireless Communications
This paper considers a passive intelligent reflecting surface (IRS)-assisted integrated radar and communication system for multi-target detection and multi-user communications. To balance the communication and sensing performance, we propose an alternating optimization algorithm to optimize the worst-case weighted sum of the radar waveform minimum mean square error (MSE) and Multiuser interference (MUI) in Communication, under the spectrum compatibility and power constraints. The proposed algorithm utilizes a novel Tchebycheff optimization framework that decomposes the multi-objective optimization problem into three subproblems by optimizing the radar transmitted sequences, communication transmitted sequences, and IRS phase configuration. We propose an alternating optimization algorithm which incorporates alternating direction penalty method (ADPM) and element-wise block coordinate descent (E-BCD) frameworks to efficiently solve the optimization problem. Extensive numerical simulations validate the effectiveness of the proposed method, demonstrating significant performance improvements in both minimizing radar MSE and communication MUI and better convergence speed.
18 October 2023 in IEEE Transactions on Circuits and Systems II: Express Briefs
With the dramatic evolution of information transportation, the RadCom system can simultaneously operate both radar and communication functionalities. In this brief, a novel scheme for the RadCom system is formulated to simultaneously consider the properties of both radar beampattern and communication signal-to-interference-plus noise ratio (SINR). The design degrees of freedom (DoFs) under control are the transmit beamformer vectors. The established formulation is non-convex with multiple designed variables. To overcome this difficulty, an alternating optimization (AO) procedure is proposed on the basis of the block coordinate descent method with sequential convex approximation (SCA-BCD), alternating direction method of multipliers during SCA procedure (SCA-ADMM), and Taylor transformation framework. The convergence of AO procedure is also proved analytically. Finally, simulation results manifest that the presented framework gains superior properties.
22 April 2024 in IEEE Transactions on Instrumentation and Measurement
This article focuses on achieving a reliable identification of herbal medicine while minimizing the impact of sensor drift. Therefore, 21 metal oxide sensors are implemented as sensitive units during the initial construction of the sensor array. Based on this, to fully exploit odor data and reduce computational burdens, we propose a weighted feature combination selection strategy that evaluates a feature set based on both label relevance and feature stability. To further enhance the long-term stability of the system and reduce the number of sensors, we introduce a combined scheme involving drift calibration and array optimization design. Specifically, we optimize a subspace projection matrix that considers minimizing the maximum mean discrepancy (MMD), preserving the Laplace manifold for the source domain, and maximizing the correlation coefficient between the source and target domains to compensate for sensor drift. Subsequently, we develop sensor array optimization using an embedded composition of feature relevancy (CFR) framework to eliminate redundant information in the target domain. Experimental results demonstrate that, compared to existing approaches, the proposed calibration framework achieves the best system performance with the minimum number of sensors.
17 July 2024 in IEEE Sensors Journal
In order to meet the demand for odor detection in herbal medicine, a multi-sensor detection system consisting of Metal Oxide Sensors (MOS) is designed. With the consideration of possible sensor drift existing in the herbal medicine datasets, a subspace projection method considering multiple constraints including the preservation of discriminative manifold is considered. Besides, Discriminant Extreme Learning Machine based on a Chebyshev Polynomial Weighting (DELM-CPW) is proposed to overcome the negative impact on the calibration recognition result due to the unbalanced number of samples of different classes in the source domain data. Moreover, the odor data of four classes of herbal medicine are collected by our system. They are analyzed with various machine learning algorithms, and the proposed algorithm obtains an accuracy of 0.8393 and an F1 score of 0.8313, which are higher than other classical algorithms, it can validate the effectiveness of the designed system and show the potential of the system to be applied in the intelligent identification of herbal medicine.
5 December 2024 in IEEE Transactions on Instrumentation and Measurement
This article designs a lung cancer odor detection system with preconcentration function, which can be used to collect exhaled gas and predict lung cancer. Combining the adsorption characteristics of the preconcentration system, we analyze and evaluate the effectiveness of the multiple phases of the experiment. An unsupervised cross-domain category difference maximization model is proposed to address the common drift problem in sensors for drift compensation. The proposed model considers the consistency of both marginal distribution and conditional distribution, improving the compensation performance of sensor drift by combining domain adaptation (DA) and Bayesian probability to maximize category information. For the information redundancy existing in sensor array, a new evaluation criterion based on mutual information (MI) is proposed, which reduces the amount of features and optimizes the sensor array by vectorizing them. The derived scheme solves the problem of both sensor drift and redundancy of array information simultaneously. The final experimental data indicate the effectiveness of the proposed design. The final experimental data indicate the effectiveness of the proposed design. The final experimental results show that the proposed modified class difference allotrope adaptation algorithm can achieve the recognition accuracy of 0.8644, which is the highest performance among the comparison algorithms, while the unmigrated data can only achieve the recognition accuracy of 0.5254. In terms of sensor optimization, the accuracy of the proposed array optimization method can reach 0.8983 on the data of two features from four sensors, and the optimal results can be achieved in the comparison of array optimization algorithms.