|
Improvement of K-Means clustering algorithm using genetic algorithm for spatial analysis of the oil spill detection in polarimetric SAR images
|
Mehrdad Kaveh   , Yasser Ebrahimian Ghajari * |
| Babol Noshirvani University of Technology |
|
|
Abstract: (771 Views) |
The existence of the oil spills at the bottom of the ocean and sea is one of main concerns for researchers in marine ecosystem terrains. In this study, K-Means clustering based on genetic algorithm (GA) has been used in order to detect the oil spills at the sea bottom. The main objective of the developed K-Means with GA is to cause an intelligent search which not only randomly determines the initial cluster centers but also tries to find out the optimal ones for the clusters. To achieve this goal, firstly the required preprocessing steps for SAR images such as radiometric correction, speckle reduction and POLSAR feature extraction were applied. Then the optimal cluster centers were determined by GA in order to have maximum extra_cluster distance. Finally, to find out the final optimal centers, K-Means algorithm was used in order to have maximum intra_cluster similarity. The ground truth digitized from Pauli RGB image of POLSAR image was utilized to evaluate the performance of the clustering methods. Furthermore, in order to evaluate the improved K-Means algorithm by GA, particle swarm optimization (PSO), biogeography-based optimization (BBO), artificial bee colony (ABC), and the standard K-Means clustering methods were used. The results gained by the improved GA-K-Means have more validity and precision than the other algorithms. The feature Entropy could obtain the overall accuracy of 83.24% that has a lower accuracy compared with the other features. But it shows a higher validity. The features ODD of Yamagouchi, ODD of Freeman decompositions and C11 of covariance matrix have reached the overall accuracy of 90% but have the noticeable values of the second type error by 18%, 11% and 12% which demonstrate the lowest validity in comparison with the other features.
|
|
| Keywords: Oil spill, SAR images, feature selection, K-Means, genetic algorithm. |
|
|
Full-Text [PDF 1809 kb]
(120 Downloads)
|
Type of Study: Research |
Subject:
GIS
Received: 2022/06/16 | Accepted: 2023/03/6 | ePublished ahead of print: 2024/08/6 | Published: 2024/10/29
|
|
|
|
|
|
|
| Send email to the article author |
|
|
