Comparing Sensors for Feature Extraction

Abstract

The extraction of features from coarse resolution satellite imagery is reliable for regional scale of the area, primarily, and applicable to a relatively smaller area to some extent. Very high-resolution (VHR) satellite imagery from space borne data is useful for small-scale regions. However, the price is very costly and depends on the cloud cover. The data required during the rainy season for flood studies cannot be used at all. On the other hand, aerial borne data is emerging to replace the space borne data due to easier accessibility, availability and also for a better resolution. The conventional methods to extract features itself have challenges and limitations due to pixel-based classification. The high spectral variability within classes reduces the accuracy of VHR image in pixel-based classification. Therefore, we used the object-based image analysis (OBIA) techniques in the study for overcoming such conventional difficulties. In order to compare the classification with a different source of data throughout the study, we have demonstrated the comparison of image classification using different sources of data for feature extraction. In this study, we used Unarmed Aerial Vehicle (UAV) data of Chingrajpara slum area in the state of Chhattisgarh, India having various morphological features. Firstly, we segmented the images of the study area to enhance the classification accuracy to compare with the results of space borne data of Mumbai slum area in the state of Maharashtra, India. We applied in different source of data to extract the features of formal buildings, vegetation, roads and informal settlements subjected to the availability of features in the subset. The result of the classification using UAV data is comparatively better and getting more than 90% accuracy as compared to the accuracy results of space borne data. Since the accuracy has been depending upon location-specific, the sensors having better classification accuracy can be suggested for further classification having same specified features in other locations. Furthermore, the results can be used for monitoring and rapid digitization purposes in digital repository and for disaster risk reduction, especially.