Early detection of high water saturation spots for landslide prediction using thermal image analysis
Landslide hazard is often discussed in electronic media and newspapers. Due to this problem, the government needs to bear millions of Malaysian ringgit to repair the infrastructures and utilities that had been ruined and to compensate the victims involved. Early warning system is one of the effectiv...
| Summary: | Landslide hazard is often discussed in electronic media and newspapers. Due to this problem, the government needs to bear millions of Malaysian ringgit to repair the infrastructures and utilities that had been ruined and to compensate the victims involved. Early warning system is one of the effective ways to reduce damage caused by landslides. Based on the literature found, there are many conventional methods to predict landslide that had been used previously such as remote sensing, wireless sensor network and many more. Basically, landslides happen due the many factors such as slope gradient factor, geological weathering and human-related activities such as deforestation. The main factor for landslide is water saturation, caused by heavy rain. Our naked eyes cannot see the water saturation in the soil. Hence, to solve this issue, this study investigates a new method to detect water saturation spots which is integrated with a thermal image camera to provide early detection of landslide. Thermal camera is selected because it provides accurate predictions on where landslides could occur. Thermal imaging is a technique that converts the invisible radiation into visible image for analysis and feature extraction. The images are processed using image processing software. Performance of image processing software is based on how accurate Region of Interest (ROI) detection is to eliminate unwanted pixels from an image. There are three segmentation algorithm used in this study which are HSV, K-Means and Feature Matching. The result reveals that HSV color space technique provides the best segmentation with average misclassification error equals to 0.00165 for abnormal images, 0.0061 for normal images and 0.0014 for combination of abnormal and normal images. Furthermore, the prediction method should make decision and classify the images into correct groups. Therefore, after the ROI has been detected, feature extraction and classification must be performed. Statistical based features namely minimum, maximum, mean and standard deviation were extracted from each image channels. The results show that the classifications using linear thresholding had sorted the image into correct group successfully. |
|---|