Date of Award
5-2024
Degree Type
Masters Thesis
Degree Name
Master of Science (MS)
School
Computing Sciences and Computer Engineering
Committee Chair
Chaoyang Zhang
Committee Chair School
Computing Sciences and Computer Engineering
Committee Member 2
Yuanyuan Zhang
Committee Member 2 School
Leadership
Committee Member 3
Zhaoxian Zhou
Committee Member 3 School
Computing Sciences and Computer Engineering
Abstract
Lack of sidewalk pavement can contribute to pedestrian fatalities and injuries in the USA. Although many researchers have conducted research on sidewalk detection, there are not many publicly available datasets that we would work on for sidewalk classification. In this study, I conducted classification tasks using pretrained CNN models including VGG16 and ResNet50. I extended these models by adding custom layers at the top of pretrained layers, employing various techniques to improve the classification accuracy.
The dataset comprises 4,731 images of sidewalk based on occlusion levels, including images where the sidewalk is visible from overhead and instances where sidewalk is occluded by tree canopies, buildings, or vehicles etc. These images were collected manually using ArcGIS Pro from Allegheny County, PA USA.
Different preprocessing techniques were applied to the dataset, such as image resizing, image extraction from bounding boxes, and image mapping to actual class label. Due to the class imbalance nature of the dataset, an augmentation technique was employed to augment the minority class and reduce the imbalanced. The data was partitioned into a training (80%) and test (20%) sets. The models were trained with and without augmentation and their performances were evaluated using metrics including precision, recall, F1- score, accuracy, and area under receiver operating curve. The results showed VGG16 outperforms ResNet50 in sidewalk classification and employing data augmentation technique to the minority class proves beneficial when dealing with imbalanced data in image classification tasks.
Copyright
2024,Subeksha Khanal
Recommended Citation
Khanal, Subeksha, "Satellite Image Analysis and Sidewalk Classification using Deep Learning Models" (2024). Master's Theses. 1041.
https://aquila.usm.edu/masters_theses/1041