ECE 697 final projects, Professor Malloy
Thursday August 5th (10AM – 12:40PM)
10:00 – 10:20: Yeongeun Kim. A Convolutional Approach to Quality Monitoring for Laser Welding. Laser Welding is a technique used to achieve successful hairpin processing in electric motor manufacturing. Alignment and tooling are the two primary difficulties encountered when laser welding hairpins. This project uses the concept of VGG block in VGG models to extract features from the result images. The project explores optimization methods for better image classification. [presentation] [codebase]
10:30 – 11:00: Parker Fortier and Matthew Henningsen. Electroencephalogram Modeling. Electroencephalograms measure the electrical activity within the brain. This project uses electroencephalographic data to model the brain as a multivariate autoregressive (MVAR) model and investigates the optimal memory parameter (model order) to be used. The memory parameter is the number of preceding observations used to predict the current output. [presentation] [codebase]
11:10 – 11:30: Jun Lin Tan. Hey Bucky. Phrase detection of the phrase ‘Hey Bucky’. [presentation][codebase]
11:40 – 12:00: Rebecca Mercer. WiFi Access Point Mapping and Embedding. This project aims to build a network of campus WiFi access points from time series data. The concept of multidimensional scaling is utilized to discover spatial relationships between access points from temporal data. [presentation][codebase]
12:10 – 12:40: Xu Han and Evan Wang. Chinese License Plate Recognition. Chinese license plate recognition is used to recognize characters, numbers, and characters in Chinese on the Chinese license plates. Image processing techniques are used to preprocess the images of Chinese license plates. Deep learning techniques are used for license plate recognition. [presentation][codebase]
Friday August 6th (10AM – 12:00PM)
10:00 – 10:20: Harry Xue. Short-Term Stock Market Prediction.While an investor would hold stocks for a long period of time, day traders seek to capture short-term swing of the stock market. Whether the market goes up or down is largely dictated by computer algorithms, so is there a way to learn its patterns? This project attempts to predict the short-term trend and price of stock market indices using neural networks and reinforcement learning. [presentation][codebase]
10:30 – 10:50: Graham Schloesser. CT Generation using CycleGAN. CycleGAN is a generative adversarial network that allows for unpaired image-to-image translation. This project explores generating synthetic CT scans from MRI inputs using cycleGAN. [presentation][codebase]
11:00 – 11:20: Elizabeth Murphy. Mileage extraction from odometer image. This project is to extract mileage from odometer images. Odometer images are preprocessed to increase the resolution and sharpness. A mobilenet object detection model is used to extract the odometer screen and then a second mobilenet object detection model is used to extract the digits from the screen. [presentation][codebase]
11:30 – 12:00: Renwen Cui and Zhenye Li. Monocular depth estimation using a deep network. Monocular depth estimation is used to reconstruct depth maps from one viewpoint. The aim of this project is to build an encoder-decoder neural network architecture with the help of a pre-trained DenseNet model to predict the depth map given a single RGB image. [presentation][codebase]