Signals & models
Wavelet denoising, windowed features, CNN training and confusion-matrix analysis.
Loading content
Curriculum vitae
Electrical engineering · Machine learning · Software & systems delivery
Tianjin, China · Updated Sep 19, 2026
M.S. in Electrical Engineering from NYU and B.S. in Electronic Information Science and Technology from Wuhan University of Technology. I apply signal processing and machine learning to sensor data, classification models and embedded systems, with experience in on-site smart-mining integration and software delivery. My interests span AI applications, software engineering and embedded electronics, grounded in debugging, experiment records and verifiable results.
Wavelet denoising, windowed features, CNN training and confusion-matrix analysis.
Bilingual content systems, Web 3D, local editing tools and automated checks.
STM32 firmware, analog-front-end simulation, field integration and user training.
M.S. in Electrical Engineering · New York, NY
B.S. in Electronic Information Science and Technology · Wuhan, China
Independent designer & developer · Tianjin
Sep 6, 2026 validation: reduced the calligraphy navigation index from 180,672 B to 78,160 B (−56.7%), with 1,425 navigation-equivalence comparisons. The same build covered 360 public localized paths and passed 422 tests.
Figures refer to that specific index and historical validation run. AI assisted development; the platform does not include validated RAG or online model-inference features.
Embedded development lead · New York
Outputs are relative indicators and states. The motion-to-stillness rule has no clinical FOG validation and does not establish disease severity or a diagnosis.
Team lead, 3 members · Performed all circuit simulations · Remote
Nominal simulation: ~45 dB gain · 0.11–249 Hz bandwidth · 4.83 μW power · 3.23 μV RMS input-referred noise.
Met the <5 μW power and <4 μV RMS noise targets. Results are LTspice circuit-level simulations, not full-device power or hardware measurements.
Undergraduate thesis · Wuhan University of Technology · Wuhan
Achieved 90.04% character accuracy on 17,757 mixed unoccluded and simulated-occlusion test images.
This is character-level classification accuracy, not full-plate recognition or system accuracy under real-world occlusion.
Project lead · Physics Innovation Laboratory, WUT · Wuhan
Historical records describe return-to-charge below 30% battery and airborne transfer experiments. The 30% value is a trigger threshold, not charging efficiency or success rate.
Sole Author · CONF-CIAP 2024 (3rd International Conference on Computing Innovation and Applied Physics)
MCC Wukan Zhicheng (Wuhan) Engineering Technology Co., Ltd. · Wuhan, China
Wuhan Research Institute of Posts and Telecommunications (Fiberhome Telecom Tech Academy) · Wuhan, China
University professional training in application and network simulation; no commercial network deployment.
Amateur Radio Association · Wuhan University of Technology · Wuhan, China