Correction of training process in object-based image interpretation via knowledge based system capabilities

For accurate interpretation of high-resolution images, correct training-samples are required to model the decision-making process for data classification. The automatic production of that is an important step; however, the control and correction, whether at the level of pruning the wrong data or resolving their defects, should be considered. Accordingly, in this research, a hybrid (combined deductive and inductive) interpretation system by modeling and improving the training processing was proposed for very high-resolution images. In other words, an automatic knowledge-based method is performed to apply the modeling of the ontological relationships in order to train and control the object-based support vector machine classification. In the correction process, sequence stages including automatic separation and expanding the training data and the ontology properties of the target classes for improving the defect of training data were used. Finally, high-resolution test images are used to validate the results and evaluate the method. In this respect, the proposed method is tested in different implementation cases and compared with other algorithms in each step. The experimental results indicate the reliability and efficiency of the proposed method.