A New Approach for Bright Circular Oil Tanks Detection in High Resolution Optical Imagery
A New Approach for Bright Circular Oil Tanks Detection in High Resolution Optical Imagery
Moein Zalpour1 Gholamreza Akbarizadeh2 Navid Alaei-Sheini3
1) Shahid Chamran University of Ahvaz,
2) Shahid Chamran University of Ahvaz,
3) Shahid Chamran University of Ahvaz,
IOI Code
:
XBFC-DDHAA
Publication :
4th International Congress on Engineering, Technology and Applied Science - Auckland University of Technology(etas2019.com)
Abstract :
With the rapid pace of technology into object detection, automatic oil tank detection is considered one of the important domains from remote sensing images. This paper proposes an efficient three-step framework to detect bright circular oil tanks: 1) ROI extraction, 2) segmentation, and 3) detection. First, a faster region-based convolutional neural network (Faster R-CNN) which is trained with labeled ground truth is used to extract oil tank depots from the input image. This step is not enough to separate objects from the background. In the second step, we utilized improved co-segmentation based on saliency co-fusion to segment out bright oil tank objects. In this step, after segmentation, a binary map of resultant helped us to find detection of desired objects. Then, in the last step, an improved Hough transform (HT) is used to detect circular shapes which are considered as oil tanks. Experiments demonstrate that the proposed framework is effective in oil tank detection methods.
Keywords : oil tank detection faster region based convolutional neural network (faster r cnn) deep learning co segmentation saliency co fusion binary map hough transform (ht)