Colorization of black-and-white aerial photographs using deep learning for object-based image analysis land use classification | |
| Author | Arunothai Waesonthea |
| Call Number | AIT Thesis no.RS-23-07 |
| Subject(s) | Land use--Classification Aerial photographs Deep learning (Machine learning) |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Remote Sensing and Geographic Information Systems |
| Publisher | Asian Institute of Technology |
| Abstract | The multispectral image comprising various spectral ranges is used to classify land use, and the first satellite multispectral image designed to study Earth started in the 1972s. Thailand has used historical black-and-white aerial photographs that started recording in 1954s to prove land use for issuing land title deeds and verifying right ownership possession. However, historical aerial photographs lack various spectral ranges because the information is mainly recorded in a panchromatic band or black-and-white color. This study uses the Pix2Pix model of a cGAN to predict color channels for an input black-and-white image. The goal is to find a suitable model to convert historical black- and-white photos into color (RGB) images. The results indicate that the U-NET network within the Pix2Pix is quite effective, with a PSNR of 33.025 and an SSIM of 0.961, indicating that the model successfully learned images. After then uses the OBIA to classify land use on black-and-white and colorized images, which is improved from the colorization model. These techniques used aerial photographs in 1954 in Maha Sarakham province, Thailand. The suitable parameter set across all classes is a scalc of 50, a shape of 0.3, and a compactness of 0.8 in the multiresolution segmentation process. As a result, land use classifications on colorized images showed significant improvements over black-and-white photos at +5%.This study concludes that the colorized image improved with the colorization model is superior to the black-and- white image for classifying land use. Particularly, vegetation groups are more accurate when enhanced images from the model are used. Therefore, this success can support the mission of investigating traces of land use in the past to consider the issuance of land title deeds and prove people of rightful ownership possession. |
| Year | 2023 |
| Type | Thesis |
| School | School of Engineering and Technology |
| Department | Department of Information and Communications Technologies (DICT) |
| Academic Program/FoS | Remote Sensing and Geographic Information Systems (RS) |
| Chairperson(s) | Sarawut Ninsawat |
| Examination Committee(s) | Tripathi, Nitin Kumar;Sanit Arunplod |
| Scholarship Donor(s) | His Majesty the King's Scholarships (Thailand) |
| Degree | Thesis (M. Sc.) - Asian Institute of Technology, 2023 |