A new method for semi-automatic classification of remotely sensed images developed based on the cognitive approaches for producing spatial data required in geomatics applications

Abstract

In many remote sensing image analysis processes, the human operator intervention is an indispensable component. One particular image analysis task where the human operator has a crucial role is in the selection of training data for supervised image classification methods in which essential training data selection is manually carried out by a human operator. This process is facilitated by the operator’s former experiences, perceptual skills, and knowledge about the area of the study. Understanding the expert’s cognitive processes, including reasoning, decision-making, and uses of knowledge to extract training data, can help improve task automation. This study regards training data extraction as a cognitive-behavioral task and attempts to extract semi-automatic training sites from an existing map based on the explanations of the expert performing the task. To support this cognitive approach, a combination of knowledge-based system, integration of remote sensing image analyses, GIS analyses, shape descriptor analyses, and artificial intelligence is used. Finally, to evaluate the reliability and accuracy of the extracted training data sites, they employed three supervised image classification algorithms, artificial neural network (ANN), support vector machine (SVM), and random forest (RF), and results were compared. Overall classification accuracy was obtained were high for three methods: ANN (90.43 %), SVM (86 %), and RF (84.09 %). Also, the kappa coefficients for each of methods are as follows: ANN (0.88), SVM (0.83), and RF (0.80). The results of this classification indicated that the extracted sample data from cognitive approach are reliable and help to semi-automation of supervised image classification process.