|
" Deep Learning for Remote Sensing Image Processing "
Lu, Yan
Li, Jiang
| Document Type
|
:
|
Latin Dissertation
|
| Language of Document
|
:
|
English
|
| Record Number
|
:
|
1054430
|
| Doc. No
|
:
|
TL53547
|
| Main Entry
|
:
|
Lu, Yan
|
| Title & Author
|
:
|
Deep Learning for Remote Sensing Image Processing\ Lu, YanLi, Jiang
|
| College
|
:
|
Old Dominion University
|
| Date
|
:
|
2020
|
| Degree
|
:
|
Ph.D.
|
| student score
|
:
|
2020
|
| Note
|
:
|
121 p.
|
| Abstract
|
:
|
Remote sensing images have many applications such as ground object detection, environmental change monitoring, urban growth monitoring and natural disaster damage assessment. As of 2019, there were roughly 700 satellites listing “earth observation” as their primary application. Both spatial and temporal resolutions of satellite images have improved consistently in recent years and provided opportunities in resolving fine details on the Earth's surface. In the past decade, deep learning techniques have revolutionized many applications in the field of computer vision but have not fully been explored in remote sensing image processing. In this dissertation, several state-of-the-art deep learning models have been investigated and customized for satellite image processing in the applications of landcover classification and ground object detection. First, a simple and effective Convolutional Neural Network (CNN) model is developed to detect fresh soil from tunnel digging activities near the U.S. and Mexico border by using pan-sharpened synthetic hyperspectral images. These tunnels’ exits are usually hidden under warehouses and are used for illegal activities, for example, by drug dealers. Detecting fresh soil nearby is an indirect way to search for these tunnels. While multispectral images have been used widely and regularly in remote sensing since the 1970s, with the fast advances in hyperspectral sensors, hyperspectral imagery is becoming popular. A combination of 80 synthetic hyperspectral channels with the original eight multispectral channels collected by the WorldView-2 satellite are used by CNN to detect fresh soil. Experimental results show that detection performance can be significantly improved by the combination of synthetic hyperspectral images with those original multispectral channels. Second, an end-to-end, pixel-level Fully Convolutional Network (FCN) model is implemented to estimate the number of refugee tents in the Rukban area near the Syrian-Jordan border using high-resolution multispectral satellite images collected by WordView-2. Rukban is a desert area crossing the border between Syria and Jordan, and thousands of Syrian refugees have fled into this area since the Syrian civil war in 2014. In the past few years, the number of refugee shelters for the forcibly displaced Syrian refugees in this area has increased rapidly. Estimating the location and number of refugee tents has become a key factor in maintaining the sustainability of the refugee shelter camps. Manually counting the shelters is labor-intensive and sometimes prohibitive given the large quantities. In addition, these shelters/tents are usually small in size, irregular in shape, and sparsely distributed in a very large area and could be easily missed by the traditional image-analysis techniques, making the image-based approaches also challenging. The FCN model is also boosted by transfer learning with the knowledge in the pre-trained VGG-16 model. Experimental results show that the FCN model is very accurate and has less than 2% of error. Last, we investigate the Generative Adversarial Networks (GAN) to augment training data to improve the training of FCN model for refugee tent detection. Segmentation based methods like FCN require a large amount of finely labeled images for training. In practice, this is labor-intensive, time consuming, and tedious. The data-hungry problem is currently a big hurdle for this application. Experimental results show that the GAN model is a better tool as compared to traditional methods for data augmentation. Overall, our research made a significant contribution to remote sensing image processing.
|
| Descriptor
|
:
|
Aerospace engineering
|
|
|
:
|
Artificial intelligence
|
|
|
:
|
Remote sensing
|
| Added Entry
|
:
|
Li, Jiang
|
| Added Entry
|
:
|
Old Dominion University
|
| |