| Author: | Shi, Yunhao |
| Title: | Deep learning based single-pixel imaging |
| Advisors: | Chen, Wen (EEE) |
| Degree: | M.Sc. |
| Year: | 2023 |
| Department: | Department of Electrical and Electronic Engineering |
| Pages: | iv, 56 pages : color illustrations |
| Language: | English |
| Abstract: | Currently, in a digital era, there is an increasing demand for low-latency, high-volume applications in industries, such as multimedia and live broadcasting. The Nyquist sampling theorem highlights a fundamental concern regarding data sampling. In order to recover the signal at the receiving end, sampling frequency must be equal to or greater than twice the highest frequency of signal spectrum. However, with a gradual expansion of the Internet of Things, the era of big data has stimulated the growth and advancement of sampling technology. Despite data is sampled, it remains extensive, posing difficulties for the internet and communication infrastructure in managing the surge of data volume, and possibly even experiencing overload. To address these issues, scholars have introduced a new data compression system, called compressed perception theory. This system seamlessly integrates space-time transformations of data sampling and information compression, surpassing the constraints of traditional sampling theorems. This breakthrough offers a fresh solution to the challenges of limited and restricted communication resources. In this dissertation, a method is presented for single-pixel imaging (SPI) achieved through extensive and comprehensive research employing deep learning techniques. In SPI, the light illuminates a series of illumination patterns onto the object, and a single-pixel detector is used to collect a series of light intensities collected at the receiving end without spatial resolution. A coarse recovered image can be obtained by using a second-order correlation algorithm, e.g., differential ghost imaging (DGI). Then, these recovered images with DGI are fed into modified deep neural network (DNN) and convolutional neural network (CNN) to learn to recover ghost images with the higher quality. Several simulations are conducted to verify the effectiveness of DNN and CNN. It is found that speckle noise in the recovered ghost images by DGI can be suppressed, and the reconstructed objects can produce superior image quality at the same compression rate by optimizing the measurement matrix and reconstruction. Compared to other methods that use artificial coding choices in compressed imaging, the designed deep neural network has the ability to learn from a large dataset of images to achieve a more objective illumination coding pattern. This, in turn, leads to an improved imaging quality at the same compression rate. |
| Rights: | All rights reserved |
| Access: | restricted access |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| 8284.pdf | For All Users (off-campus access for PolyU Staff & Students only) | 3.65 MB | Adobe PDF | View/Open |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13877