/
Languages
English
Japanese
Social networks
Skills
Creativity
Interpersonal Skills
Problem-Solving
Leadership
Teamwork
Nguyen Duy Kha
AI Researcher, AI Engineer, Python web programming Take advantage of skills and understanding of technology along with experiences through AI and web projects with the hope of creating innovative products that benefit life and bring more value to the community. Education
Research experiences
Work experience
nguyenduykha12762@gma
il.com
Ninh Kieu, Can Tho
Date of birth 26/08/2000
https://github.com/Nguyen
DuyKha/
Vietnamese
@duykhanguyen2000
IT Engineer
Since August 2018 Can Tho University Can Tho
In 2019, Certificate of Completion Information Security Basic 1 workshop Excellent student in the school year 2019-2020
Excellent student in the school year 2020-2021
Register and manage Student of 5 merits
From April 2021 to September 2021 Can Tho University Can Tho Lead researcher of Scientific Research
Build websites with Django
Document Oriented Database MongoDB
Deploy on the Heroku platform
Internships
From May 2021 to July 2021 DELTA BRAINS JSC Can Tho In order to solve difficulties and apply artificial intelligence in life, building a website to extract information from Japanese health insurance cards will bring many conveniences, which will help users easily extract export personal information from the image of the Japanese health insurance card. That is also the reason why I carried out the topic: "Building a website to extract information from Japanese health insurance cards".
Learn about OpenCV library in C#
Learn about Tesseract.Net SDK library
Learn about RESTful API
Build APS.NET WebAPI to extract information from Japanese health insurance cards with Tesseract.Net SDK library
Build website interface with Angular framework
Learn about Microsoft Azure Computer Vision's artificial intelligence support service
Integrate more API to extract information from Japanese health insurance cards through OCRSpace service
/
Programing skills
C++, C#
C, Java
Python
Web Development
Personal projects
Object Detection with Django
From April 2020 to July 2020
A web-app that provides object detection using YOLOv3. The website shows the detection output with bounding boxes around the detected objects. There will be no box if the input doesn't contain any object. YOLO trained on the COCO dataset consists of 80 labels.
Face Mask Detector
From June 2020 to August 2020
Using a dataset of 678 images captured in public places, Convolutional Neural Network and OpenCV preprocessing were used to improve the performance of transmitted datasets. Object detection problem using Yolov3 model to identify objects in real-time with 3 labels (layers) as follows: mask = face with the mask, none = yet define, and no-mask = not wearing a mask. Train the Yolov3 model using the Darknet-53 framework on Google Colab. Research results with FPS = 1.72 (Frames Per Second) on the data set for physical devices only CPU. In the future, the research will expand the development direction on more specialized physical devices and the more extensive data set helps the system to have the speed of identifying objects closer to real-time, with higher accuracy. Chatbot
From January 2021 to February 2021
Build a chatbot using deep learning techniques. The chatbot will be trained on the dataset which contains categories (intents), patterns and responses. Let’s create a retrieval based chatbot using NLTK, Keras, Python, etc Building a model to design sneaker using DCGAN
From June 2021 to August 2021
In this study, a dataset of 12856 images of trainers from the Sneakers and Athletic Shoes section of the UT Zappos50K dataset is used, through images preprocessing by transformations of the PyTorch library, which produces the images whose CHW size are 3 64 64. Then, applying the DCGAN model with 2 labels (classes) as follows: 0 = fake image and 1 = real image, combine the technique of freezing discriminator model when the conditions are met. The results obtained with the loss function value BCELoss of generator and discriminator on the dataset are 1.6367 and 1.1280 respectively. Real-time Sudoku puzzle game on mobile
From August 2021 to October 2021
This study proposes a method using visual techniques to simultaneously digitally detect and decrypt a Sudoku puzzle and solve the puzzle using a BFS algorithm. The application developed with the intention to realize any sudoku puzzle in real time through the camera from the mobile device. Initially, using preprocessing algorithms such as: HOG transform, geometric transform, ... to create a premise to recognize (distinguish) the digits from the SVM model based on their pixel positions in the image. Then, they are stored in the corresponding positions in the 9 9 matrix and the algorithm is applied. Finally, perform virtualization of the results onto the mobile device screen. Experimental results show that this method has an average accuracy of the SVM classification model is 99.14% with two separate data sets, in which, the training set consists of 5280 images and the test set is 2640 images.