I am a Master Student of the TU Braunschweig (TU9 Universities), Germany.
I got my bachelor's degree of communication engineering in the Xidian University of China.
In this paper we describe several light-weight networks based on MobileNetV2, ShuffleNet, Mixed-scale DenseNet designed for semantic image segmentation task. Additionally, we introduce GAN for data augmentation (pix2pixHD), concurrent Spatial-Channel Sequeeze & Excitation (SCSE) and Receptive Field Block (RFB) to the proposed network. We measure our performance on Cityscapes pixel-level segmentation, and achieve up to 70.72% class mIoU.
In this work we propose two postprocessing approaches applying convolutional neural networks (CNNs) either in the time domain or the cepstral domain to enhance the coded speech without any modification of the codecs. The time domain approach follows an end-to-end fashion, while the cepstral domain approach uses analysis-synthesis with cepstral domain features.
Left turns may generate efficiency problems which can be solved by appropriately prohibiting left turns. The goal of this paper is to propose a method for purpose of minimizing total travel times in urban road networks by prohibiting left turns. With left turn prohition, the signal timing plan is optimized with the lane-based method because the method can properly handle both signal timing optimization and lane assignment. The total travel time is calculated with link flows and link travel time being estimated with signal settings.
Speech prediction plays a key role in many speech signal processing and speech communication methods. While linear prediction of speech is well-studied, nonlinear speech prediction increasingly receives interest especially with the vast amount of new neural network topologies proposed recently. In this paper, nonlinear speech prediction is conducted by a special kind of recurrent neural network not requiring any training beforehand, the echo state network, which adaptively updates its output layer weights. Simulations show its superior performance compared to other well-known prediction approaches in terms of the prediction gain, exceeding all baselines in all conditions by up to 8 dB.
Mask-YOLO: Efficient Instance-level Segmentation Network based on YOLO-V2 (Excellent Intern of Intel).
Technical Features of Mask-YOLO:
1. Multiple scales: Encoder-Decoder.
2. Learnable attention unit.
3. Cascade label guidance.
Investigation of Acoustic Echo Cancellation (AEC) and Trying to use Recurrent Neural Network/Echo State Network to do Nonlinear and Linear Acoustic Echo Cancellation.
RNN/ESN is good at Nonlinear AEC.
GWMS is a software package developed by me to help man to remember the german word.
It was developed based on the cross-platform C++ GUI application development framework Qt.
It contains five sub-software: GWMS, WTEditor, Forvo Realspeak Downloador, AnsleDictionary and DehelperBackupAnalyzer.
PTMS is is a system for remote monitoring of the patient's ECG, blood pressure, pulse oximetry.
Remote mobile client uses ARM11&MUC51 development board and WinCE system.
The server uses the windows system, and the communication methode are SMS and TCP.
All software are developmented using C++/ MFC.
This STM32 Development Board was developed by me and my partner on the basis of other's development boards.
The Printed Wire Board/PCB was drawn using Altium Designer by me when I was writing my bachelor's Thesis.
MSDO is a simple Oscilloscope software developed by me using Qt when I was doing my Bachelor Graduation Project.
It can receive data of fixed form via the serial port and display the waveform of the data in real time.
SMS GPRS Demo in Windows and Windows CE using MFC C++, Tested using NeoWay GPRS Module.
TCP UDP Demo in Windows and Windows CE using MFC C++, Tested in Windows and Win CE.
Matrix Calculator is a simple matrix calculator to do some linear algebra computations.
It was developed using C ++/MFC .
Using Atmega12864 MCU to create make a tuner and metronome.
It was a team project, my work in this project is to use the FFT to convert the acquired signal to the frequency domain.
The ADC/DAC experiment designed for the STM32 development board developed by myself.
All the code is written in C language.
All Other Projects In My GitHub, there have different kinds of projects using C/C++, Java, Matlab and so on.
They are WordsThesaurusEditor, AnsleDict, ForvoRealSpeakDownload, NNAEC, QtSpeechTest...
In the Bachelor stage, I mainly engaged in embedded hardware and software system development.
In the Master stage, I am mainly specialized in software development, speech signal processing, computer vision, machine learning and deep learning.
I can skillfully use a variety of programming languages, including C/C++, Java, Matlab, Python, HTML...
I have ever done some database (SQLite) related project.
I have a complete knowledge reserves of speech processing, transmission and recognition.
Application of deep learning and online learning in speech processing were my former research focus.
In the last year, I have been doing some things about machine learning, especially online learning and deep learning.
I usually use Matlab and Python to do machine learning and deep learning, especially for computer vision.
I fond of embedded system development, my once used microcontroller: MCU51, STM32, Atmega12864, ARM11.
C and C++ are my favorite programming language when doing embedded system development.
Scientific research is my dream, I like research work.
When I have a new idea, I even can be too excited to forget to sleep.
| Skills | Description | Grade |
|---|---|---|
| C/C++ | Favorite, most commonly used. | 90+ |
| Qt-C++/Jambi/MFC | Favorite, most commonly used. | 90+ |
| Java | Most commonly used. | 85+ |
| Matlab | Most commonly used. | 85+ |
| Python | Most commonly used. | 70+ |
| C♯/Xamarin | Now less used. | 50+ |
| HTML/HTML5/JavaScript | Now less used. | 50+ |
| Altium Designer | Now less used. | 50+ |