Rule-based land cover classification from very high-resolution satellite image with multiresolution segmentation

Abstract

Multiresolution segmentation and rule-based classification techniques are used to classify objects from very high-resolution satellite images of urban areas. Custom rules are developed using different spectral, geometric, and textural features with five scale parameters, which exploit varying classification accuracy. Principal component analysis is used to select the most important features out of a total of 207 different features. In particular, seven different object types are considered for classification. The overall classification accuracy achieved for the rule-based method is 95.55% and 98.95% for seven and five classes, respectively. Other classifiers that are not using rules perform at 84.17% and 97.3% accuracy for seven and five classes, respectively. The results exploit coarse segmentation for higher scale parameter and fine segmentation for lower scale parameter. The major contribution of this research is the development of rule sets and the identification of major features for satellite image classification where the rule sets are transferable and the parameters are tunable for different types of imagery. Additionally, the individual objectwise classification and principal component analysis help to identify the required object from an arbitrary number of objects within images given ground truth data for the training.

WorldView-2 Data for Hierarchical Object-Based Urban Land Cover Classification in Kigali: Integrating Rule-Based Approach with Urban Density and Greenness Indices

“…The study pointed out that the synergy between a robust classifier, such as SVM, and the integration of a geometric rule-set and the proposed density indices (UDI and GDI), is a reliable method to improve the urban land cover classification in complex urban environments. The findings in the present study concur with previous studies e.g., [25,[73][74][75], where the rule-based approach using geometric features, texture measurements and the spectral band threshold were found useful for land cover classification enhancement. Some of the features to ingest in the feature space include the bands' mean and standard deviation, and in particular the geometric features related to the object's extent and shape, such as compactness, asymmetry and rectangular fit, area, width and length.…”

Delineation of the Urban-Rural Boundary through Data Fusion: Applications to Improve Urban and Rural Environments and Promote Intensive and Healthy Urban Development

“…Therefore, researchers have developed a tool named estimation of scale parameters to quickly determine the appropriate scale parameters for image segmentation. The scale parameters determined through this tool improve the efficiency and accuracy of image interpretation [ 55 ].…”

Identifying Generalizable Image Segmentation Parameters for Urban Land Cover Mapping through Meta-Analysis and Regression Tree Modeling

“…However, in many cases, it is desirable to produce a multi-class land cover map (rather than separate maps of each individual land cover class). To use our approach for multi-class land cover map production, one possibility would be to perform classification in a step-wise manner (e.g., classifying the land cover types with lower SP values first, or vice versa), as has been done in several other GEOBIA studies that utilized multiple segmentation levels for classification [37,[54][55][56]. More sophisticated solutions for multi-scale classification also exist (and may indeed work better than our simple example), but a deeper discussion/comparison of these methods is not provided here because it is outside the main focus of our study.…”