University of Hawaii

Electrical Engineering

Machine Learning for Digital Image Processing

Date: 2018-07-12           Add to Google Calendar
Time: 9:30am
Location: Holmes Hall 389
Speaker: Evan Law, EE MS Candidate


With the rapid growth of computing capacity, machine learning (ML) algorithms have been successfully applied for digital image processing with various goals. In this work, we discuss the use of supervised machine learning methods to detect and classify infrastructure and biological specimens using images obtained with Unmanned Aerial Vehicles (UAVs). Neural Networks are typical ML solutions and consist of interconnected neurons that process the inputs to obtain desired outputs, with various hidden layers that specialize in detecting features. Convolutional Neural Networks (CNNs) are a type a Neural Networks that specialize in image analysis for classification, object detection, segmentation and image processing. Transfer Learning using existing CNNs provides quick results in image analysis without the large resources required to train a CNN from scratch. Haar Cascades, Inception, You Only Look Once (YOLO) and Object Detection API are various models explored to identify and compare images. Existing solutions requires specialized onboard hardware on UAVs, scaled CNNs, and simplified image analysis on input data.