Jiangshan Zhang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jiangshan Zhang has a total of 28 co-authors. Here are the top co-authors by number of shared publications:
- Deming Liu
Wuhan National Laboratory for Optoelectronics and School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan, China
5 shared publications - Ping Lu
Wuhan National Laboratory for Optoelectronics (WNLO) and National Engineering Research Center of Next Generation Internet Access System, School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan, China
5 shared publications - Kai Chen
Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China
2 shared publications - Qiang Wu
Northumbria University
2 shared publications - Wanjin Zhang
Wuhan National Laboratory for Optoelectronics (WNLO) and National Engineering Laboratory for Next Generation Internet Access System, School of Optical and Electronic Information, Huazhong University of Science and Technology, Wuhan 430074, China
2 shared publications - Yufeng Pan
College of Marine Geosciences Ocean University of China Qingdao China
2 shared publications
Latest Publications
- Open Access
- 11 December 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Privacy and Security
- Vol. 28 (1) , 1-27
Adversarial Attack and Defense for Commercial Black-box Chinese-English Speech Recognition Systems
The attacker can generate adversarial examples (AEs) to stealthily mislead automatic speech recognition (ASR) models, raising significant concerns about the security of intelligent voice control (IVC) devices. Existing adversarial attacks mainly generate AEs to mislead ASR models to output specific target English commands (e.g., open the door). However, it remains unknown whether AEs can be used to issue commands in other languages to attack commercial black-box ASR models. In this article, taking Chinese phrases (e.g., 支付宝付款) and “Chinese–English code-switching” phrases (e.g., 关闭GPS) as the target commands, we propose adversarial attacks for commercial multilingual ASR models. In particular, if a multilingual speech recognition model can recognize Chinese and English, we call it a Chinese–English speech recognition model. In English, the meaning of “支付宝付款” and “关闭GPS” are “Alipay payment” and “turn off GPS”, respectively. In detail, we generate transferable AEs based on the open-sourced conventional DataTang Mandarin ASR model. Given 55 target commands, the success rate for generating AEs of them is up to 96% and 80% for Aliyun ASR API and Tencentyun ASR API, respectively. Our AEs can trigger actual attack actions on voice assistants (e.g., Apple Siri, Xiaomi Xiaoaitongxue) or spread malicious messages through ASR API services, while the target commands in the AEs are inaudible to human beings. 1 Finally, by analyzing the spectrum differences between benign audio clips and AEs, we propose a general defense against adversarial audio attacks.
- 11 December 2024
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Privacy and Security
- Vol. 28 (1) , 1-27
Open AccessAdversarial Attack and Defense for Commercial Black-box Chinese-English Speech Recognition Systems
The attacker can generate adversarial examples (AEs) to stealthily mislead automatic speech recognition (ASR) models, raising significant concerns about the security of intelligent voice control (IVC) devices. Existing adversarial attacks mainly generate AEs to mislead ASR models to output specific target English commands (e.g., open the door). However, it remains unknown whether AEs can be used to issue commands in other languages to attack commercial black-box ASR models. In this article, taking Chinese phrases (e.g., 支付宝付款) and “Chinese–English code-switching” phrases (e.g., 关闭GPS) as the target commands, we propose adversarial attacks for commercial multilingual ASR models. In particular, if a multilingual speech recognition model can recognize Chinese and English, we call it a Chinese–English speech recognition model. In English, the meaning of “支付宝付款” and “关闭GPS” are “Alipay payment” and “turn off GPS”, respectively. In detail, we generate transferable AEs based on the open-sourced conventional DataTang Mandarin ASR model. Given 55 target commands, the success rate for generating AEs of them is up to 96% and 80% for Aliyun ASR API and Tencentyun ASR API, respectively. Our AEs can trigger actual attack actions on voice assistants (e.g., Apple Siri, Xiaomi Xiaoaitongxue) or spread malicious messages through ASR API services, while the target commands in the AEs are inaudible to human beings. 1 Finally, by analyzing the spectrum differences between benign audio clips and AEs, we propose a general defense against adversarial audio attacks.
- Open Access
- 5 August 2024
- Published by Elsevier in Photoacoustics
- Vol. 39, 100639
Open-closed single-tube on-beam tuning-fork-enhanced fiber-optic photoacoustic spectroscopy
- 5 August 2024
- Published by Elsevier in Photoacoustics
- Vol. 39, 100639
Open AccessOpen-closed single-tube on-beam tuning-fork-enhanced fiber-optic photoacoustic spectroscopy
- Open Access
- 29 May 2024
- Published by Elsevier in Photoacoustics
- Vol. 38, 100624
Differential photoacoustic spectroscopy for flow gas detection based on single microphone
- 29 May 2024
- Published by Elsevier in Photoacoustics
- Vol. 38, 100624
Open AccessDifferential photoacoustic spectroscopy for flow gas detection based on single microphone
- 1 May 2024
- Published by Optica Publishing Group in Photonics Research
- Vol. 12 (5) , 969-978
Miniaturized and highly sensitive fiber-optic Fabry–Perot sensor for mHz infrasound detection
Infrasound detection is important in natural disasters monitoring, military defense, underwater acoustic detection, and other domains. Fiber-optic Fabry–Perot (FP) acoustic sensors have the advantages of small structure size, long-distance detection, immunity to electromagnetic interference, and so on. The size of an FP sensor depends on the transducer diaphragm size and the back cavity volume. However, a small transducer diaphragm size means a low sensitivity. Moreover, a small back cavity volume will increase the low cut-off frequency of the sensor. Hence, it is difficult for fiber-optic FP infrasound sensors to simultaneously achieve miniaturization, high sensitivity, and extremely low detectable frequency. In this work, we proposed and demonstrated a miniaturized and highly sensitive fiber-optic FP sensor for mHz infrasound detection by exploiting a Cr-Ag-Au composite acoustic-optic transducer diaphragm and a MEMS technique-based spiral micro-flow hole. The use of the spiral micro-flow hole as the connecting hole greatly reduced the volume of the sensor and decreased the low-frequency limit, while the back cavity volume was not increased. Combined with the Cr-Ag-Au composite diaphragm, a detection sensitivity of −123.19 dB re 1 rad/μPa at 5 Hz and a minimum detectable pressure (MDP) of 1.2 mPa/Hz1/2 at 5 Hz were achieved. The low detectable frequency can reach 0.01 Hz and the flat response range was 0.01–2500 Hz with a sensitivity fluctuation of ±1.5 dB. Moreover, the size of the designed sensor was only 12 mm×Φ12.7 mm. These excellent characteristics make the sensor have great practical application prospects.
- 1 May 2024
- Published by Optica Publishing Group in Photonics Research
- Vol. 12 (5) , 969-978
Miniaturized and highly sensitive fiber-optic Fabry–Perot sensor for mHz infrasound detection
Infrasound detection is important in natural disasters monitoring, military defense, underwater acoustic detection, and other domains. Fiber-optic Fabry–Perot (FP) acoustic sensors have the advantages of small structure size, long-distance detection, immunity to electromagnetic interference, and so on. The size of an FP sensor depends on the transducer diaphragm size and the back cavity volume. However, a small transducer diaphragm size means a low sensitivity. Moreover, a small back cavity volume will increase the low cut-off frequency of the sensor. Hence, it is difficult for fiber-optic FP infrasound sensors to simultaneously achieve miniaturization, high sensitivity, and extremely low detectable frequency. In this work, we proposed and demonstrated a miniaturized and highly sensitive fiber-optic FP sensor for mHz infrasound detection by exploiting a Cr-Ag-Au composite acoustic-optic transducer diaphragm and a MEMS technique-based spiral micro-flow hole. The use of the spiral micro-flow hole as the connecting hole greatly reduced the volume of the sensor and decreased the low-frequency limit, while the back cavity volume was not increased. Combined with the Cr-Ag-Au composite diaphragm, a detection sensitivity of −123.19 dB re 1 rad/μPa at 5 Hz and a minimum detectable pressure (MDP) of 1.2 mPa/Hz1/2 at 5 Hz were achieved. The low detectable frequency can reach 0.01 Hz and the flat response range was 0.01–2500 Hz with a sensitivity fluctuation of ±1.5 dB. Moreover, the size of the designed sensor was only 12 mm×Φ12.7 mm. These excellent characteristics make the sensor have great practical application prospects.
- Open Access
- 15 November 2023
- Published by Association for Computing Machinery (ACM)
- p. 431-445
Good-looking but Lacking Faithfulness: Understanding Local Explanation Methods through Trend-based Testing
While enjoying the great achievements brought by deep learning (DL), people are also worried about the decision made by DL models, since the high degree of non-linearity of DL models makes the decision extremely difficult to understand. Consequently, attacks such as adversarial attacks are easy to carry out, but difficult to detect and explain, which has led to a boom in the research on local explanation methods for explaining model decisions. In this paper, we evaluate the faithfulness of explanation methods and find that traditional tests on faithfulness encounter the random dominance problem, i.e., the random selection performs the best, especially for complex data. To further solve this problem, we propose three trend-based faithfulness tests and empirically demonstrate that the new trend tests can better assess faithfulness than traditional tests on image, natural language and security tasks. We implement the assessment system and evaluate ten popular explanation methods. Benefiting from the trend tests, we successfully assess the explanation methods on complex data for the first time, bringing unprecedented discoveries and inspiring future research. Downstream tasks also greatly benefit from the tests. For example, model debugging equipped with faithful explanation methods performs much better for detecting and correcting accuracy and security problems.
- 15 November 2023
- Published by Association for Computing Machinery (ACM)
- p. 431-445
Open AccessGood-looking but Lacking Faithfulness: Understanding Local Explanation Methods through Trend-based Testing
While enjoying the great achievements brought by deep learning (DL), people are also worried about the decision made by DL models, since the high degree of non-linearity of DL models makes the decision extremely difficult to understand. Consequently, attacks such as adversarial attacks are easy to carry out, but difficult to detect and explain, which has led to a boom in the research on local explanation methods for explaining model decisions. In this paper, we evaluate the faithfulness of explanation methods and find that traditional tests on faithfulness encounter the random dominance problem, i.e., the random selection performs the best, especially for complex data. To further solve this problem, we propose three trend-based faithfulness tests and empirically demonstrate that the new trend tests can better assess faithfulness than traditional tests on image, natural language and security tasks. We implement the assessment system and evaluate ten popular explanation methods. Benefiting from the trend tests, we successfully assess the explanation methods on complex data for the first time, bringing unprecedented discoveries and inspiring future research. Downstream tasks also greatly benefit from the tests. For example, model debugging equipped with faithful explanation methods performs much better for detecting and correcting accuracy and security problems.
- Open Access
Optical Fiber Sensor for Curvature and Temperature Measurement Based on Anti-Resonant Effect Cascaded with Multimode Interference
In this paper, a novel inline optical fiber sensor for curvature and temperature measurement simultaneously has been proposed and demonstrated, which can measure two parameters with very little crosstalk. Two combinational mechanisms of anti-resonant reflecting optical waveguide and inline Mach–Zehnder interference structure are integrated into a 3 mm-long single hole twin suspended core fiber (SHTSCF). The 85 μm hole core gives periodic several dominant resonant wavelengths in the optical transmission spectrum, acting as the anti-resonant reflecting optical waveguide (ARROW). The modes in two suspended cores and the cladding form the comb pattern. Reliable sensor sensitivity can be obtained by effective experiments and demodulation. Through intensity demodulation of the selected dip of Gaussian fitting, the curvature sensitivity can be up to −7.23 dB/m−1. Through tracking the MZI dip for wavelength demodulation, the temperature sensitivity can be up to 28.8 pm/°C. The sensor is simple in structure, compact, and has good response, which can have a bright application in a complex environment.
Open AccessOptical Fiber Sensor for Curvature and Temperature Measurement Based on Anti-Resonant Effect Cascaded with Multimode Interference
In this paper, a novel inline optical fiber sensor for curvature and temperature measurement simultaneously has been proposed and demonstrated, which can measure two parameters with very little crosstalk. Two combinational mechanisms of anti-resonant reflecting optical waveguide and inline Mach–Zehnder interference structure are integrated into a 3 mm-long single hole twin suspended core fiber (SHTSCF). The 85 μm hole core gives periodic several dominant resonant wavelengths in the optical transmission spectrum, acting as the anti-resonant reflecting optical waveguide (ARROW). The modes in two suspended cores and the cladding form the comb pattern. Reliable sensor sensitivity can be obtained by effective experiments and demodulation. Through intensity demodulation of the selected dip of Gaussian fitting, the curvature sensitivity can be up to −7.23 dB/m−1. Through tracking the MZI dip for wavelength demodulation, the temperature sensitivity can be up to 28.8 pm/°C. The sensor is simple in structure, compact, and has good response, which can have a bright application in a complex environment.
- Open Access
Large-Dynamic-Range and High-Stability Phase Demodulation Technology for Fiber-Optic Michelson Interferometric Sensors
A large-dynamic-range and high-stability phase demodulation technology for fiber-optic Michelson interferometric sensors is proposed. This technology utilizes two output signals from a 2 × 2 fiber-optic coupler, the interferometric phase difference of which is π. A linear-fitting trigonometric-identity-transformation differential cross-multiplication (LF-TIT-DCM) algorithm is used to interrogate the phase signal from the two output signals from the coupler. The interferometric phase differences from the two output signals from the 2 × 2 fiber-optic couplers with different coupling ratios are all equal to π, which ensures that the LF-TIT-DCM algorithm can be applied perfectly. A fiber-optic Michelson interferometric acoustic sensor is fabricated, and an acoustic signal testing system is built to prove the proposed phase demodulation technology. Experimental results show that excellent linearity is observed from 0.033 rad to 3.2 rad. Moreover, the influence of laser wavelength and optical power is researched, and variation below 0.47 dB is observed at different sound pressure levels (SPLs). Long-term stability over thirty minutes is tested, and fluctuation is less than 0.36 dB. The proposed phase demodulation technology obtains large dynamic range and high stability at rather low cost.
Open AccessLarge-Dynamic-Range and High-Stability Phase Demodulation Technology for Fiber-Optic Michelson Interferometric Sensors
A large-dynamic-range and high-stability phase demodulation technology for fiber-optic Michelson interferometric sensors is proposed. This technology utilizes two output signals from a 2 × 2 fiber-optic coupler, the interferometric phase difference of which is π. A linear-fitting trigonometric-identity-transformation differential cross-multiplication (LF-TIT-DCM) algorithm is used to interrogate the phase signal from the two output signals from the coupler. The interferometric phase differences from the two output signals from the 2 × 2 fiber-optic couplers with different coupling ratios are all equal to π, which ensures that the LF-TIT-DCM algorithm can be applied perfectly. A fiber-optic Michelson interferometric acoustic sensor is fabricated, and an acoustic signal testing system is built to prove the proposed phase demodulation technology. Experimental results show that excellent linearity is observed from 0.033 rad to 3.2 rad. Moreover, the influence of laser wavelength and optical power is researched, and variation below 0.47 dB is observed at different sound pressure levels (SPLs). Long-term stability over thirty minutes is tested, and fluctuation is less than 0.36 dB. The proposed phase demodulation technology obtains large dynamic range and high stability at rather low cost.
- Open Access
Passive Homodyne Phase Demodulation Technique Based on LF-TIT-DCM Algorithm for Interferometric Sensors
A passive homodyne phase demodulation technique based on a linear-fitting trigonometric-identity-transformation differential cross-multiplication (LF-TIT-DCM) algorithm is proposed. This technique relies on two interferometric signals whose interferometric phase difference is odd times of π. It is able to demodulate phase signals with a large dynamic range and wide frequency band. An anti-phase dual wavelength demodulation system is built to prove the LF-TIT-DCM algorithm. Comparing the traditional quadrature dual wavelength demodulation system with an ellipse fitting DCM (EF-DCM) algorithm, the phase difference of two interferometric signals of the anti-phase dual wavelength demodulation system is set to be π instead of π/2. This technique overcomes the drawback of EF-DCM—that it is not able to demodulate small signals since the ellipse degenerates into a straight line and the ellipse fitting algorithm is invalidated. Experimental results show that the dynamic range of the proposed anti-phase dual wavelength demodulation system is much larger than that of the traditional quadrature dual wavelength demodulation system. Moreover, the proposed anti-phase dual wavelength demodulation system is hardly influenced by optical power, and the laser wavelength should be strictly limited to lower the reference error.
Open AccessPassive Homodyne Phase Demodulation Technique Based on LF-TIT-DCM Algorithm for Interferometric Sensors
A passive homodyne phase demodulation technique based on a linear-fitting trigonometric-identity-transformation differential cross-multiplication (LF-TIT-DCM) algorithm is proposed. This technique relies on two interferometric signals whose interferometric phase difference is odd times of π. It is able to demodulate phase signals with a large dynamic range and wide frequency band. An anti-phase dual wavelength demodulation system is built to prove the LF-TIT-DCM algorithm. Comparing the traditional quadrature dual wavelength demodulation system with an ellipse fitting DCM (EF-DCM) algorithm, the phase difference of two interferometric signals of the anti-phase dual wavelength demodulation system is set to be π instead of π/2. This technique overcomes the drawback of EF-DCM—that it is not able to demodulate small signals since the ellipse degenerates into a straight line and the ellipse fitting algorithm is invalidated. Experimental results show that the dynamic range of the proposed anti-phase dual wavelength demodulation system is much larger than that of the traditional quadrature dual wavelength demodulation system. Moreover, the proposed anti-phase dual wavelength demodulation system is hardly influenced by optical power, and the laser wavelength should be strictly limited to lower the reference error.
- 1 June 2014
- Published by Elsevier in Optics Communications
- Vol. 320, 33-37
A single-mode fiber sensor based on core-offset inter-modal interferometer
- 1 June 2014
- Published by Elsevier in Optics Communications
- Vol. 320, 33-37
A single-mode fiber sensor based on core-offset inter-modal interferometer