[Home ] [Archive]   [ فارسی ]  
:: Main :: About :: Current Issue :: Archive :: Search :: Submit :: Contact ::
Main Menu
Journal Information::
Articles archive::
For Authors::
For Reviewers::
Contact us::
Site Facilities::
Search in website

Advanced Search
Receive site information
Enter your Email in the following box to receive the site news and information.
:: Volume 10, Issue 4 (3-2023) ::
jgit 2023, 10(4): 87-108 Back to browse issues page
Transfer Learning Framework for Semantic Segmentation of High-Resolution UAV-based images in Urban Area
Abbas Majidizadeh , Hadiseh Hasani * , Marzieh Jafari
Tafresh University
Abstract:   (1838 Views)
Semantic segmentation technique for Unmanned Aerial Vehicle (UAV) data processing has been one of the leading researches in photogrammetry, remote sensing, and computer vision in recent years. This technique has attracted increasing attention from industry and academia (a wide range of academic and real-world applications). Many applications, including aerial mapping of urban scenes, positioning objects in aerial images, automatic extraction of buildings from remote sensing or high-resolution aerial images, etc., require accurate and efficient segmentation algorithms. However, proper and accurate semantic segmentation using a deep learning approach (overall training of a deep neural network with random weighting) requires a large amount of training and labeled images. As we are facing the challenge of a lack of labeled data in the field of urban aerial images, we used the Transfer Learning Approach for the semantic segmentation of the UAV-based images of urban areas in this paper. The proposed method implements a transfer learning framework based on DeepLabV3Plus convolutional encoder-decoder architecture with ResNet-50 pre-trained model in ImageNet collection for semantic segmentation of the urban scenes. The dataset studied in this research is the UAVid2020, an urban UAV-based semantic segmentation dataset from the International Society for Photogrammetry and Remote Sensing (ISPRS). We used traditional deep learning models (U-Net and Seg-Net convolutional encoder-decoder neural networks) to evaluate the semantic segmentation performance of the proposed method. Finally, the results of  the semantic segmentation of UAV-based images show the effectiveness of the proposed transfer learning framework compared to  the deep learning models, in terms of the overall accuracy metric. The DeepLabV3Plus-ResNet50 architecture achieved the best result with 81.93% compared to U-Net and Seg-Net neural networks with 74.35% and 79.15% respectively.
Keywords: Semantic Segmentation, Unmanned Aerial Vehicle (UAV), Transfer Learning, Convolutional Encoder-Decoder Deep Neural Network, DeepLabV3Plus
Full-Text [PDF 2434 kb]   (468 Downloads)    
Type of Study: Research | Subject: Aerial Photogrammetry
Received: 2022/12/28 | Accepted: 2023/05/13 | ePublished ahead of print: 2023/05/15 | Published: 2023/05/21
Send email to the article author

XML   Persian Abstract   Print

Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

Majidizadeh A, Hasani H, Jafari M. Transfer Learning Framework for Semantic Segmentation of High-Resolution UAV-based images in Urban Area. jgit 2023; 10 (4) :87-108
URL: http://jgit.kntu.ac.ir/article-1-904-en.html

Rights and permissions
Creative Commons License This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
Volume 10, Issue 4 (3-2023) Back to browse issues page
نشریه علمی-پژوهشی مهندسی فناوری اطلاعات مکانی Engineering Journal of Geospatial Information Technology
Persian site map - English site map - Created in 0.05 seconds with 36 queries by YEKTAWEB 4657