Automated vehicle detection in satellite images using deep learning | ||||
| International Conference on Aerospace Sciences and Aviation Technology | ||||
| Article 29, Volume 18, Issue 18, April 2019, Page 1-8 PDF (817.32 K) | ||||
| Document Type: Original Article | ||||
| DOI: 10.1088/1757-899X/610/1/012027 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
| Ahmad Mansour; Ahmed Hassan; Wessam M Hussein; Ehab Said | ||||
| Department of Mechatronic Engineering, Military Technical College, 11766, Cairo, Egypt. | ||||
| Abstract | ||||
| Automatic detection of small objects such as vehicles in satellite images is a very challenging task, due to the complexity of the background, vehicles colors, the large size of ground sample distance (GSD) for satellite images and jamming caused by buildings and trees. Many methods were proposed for this task by using handcrafted features (such as a Histogram of an Oriented Gradient, Local Binary Pattern, Scale-Invariant Feature Transform, etc.) along with support vector machine classifier, however, Convolutional Neural Networks (CNN) have proved to be potentially more effective. In this paper, we use two advanced deep learning frameworks, Faster Region CNN (Faster R-CNN) and Single Shot Multi-Box (SSD) based on (CNN) with Inception-V2 as a feature map generator instead of VGG-16, to detect vehicles through Transfer Learning, and making an experimental analysis comparison between the two models. Experimental results on the test dataset demonstrate the effectiveness and efficiency of the proposed methods. | ||||
|
Statistics Article View: 390 PDF Download: 232 |
||||
