In this paper, a new fuzzy rough set (FRS) method was proposed for extracting rules from an adaptive neuro-fuzzy inference system (ANFIS)-based classification procedure in order to select the optimum features. The proposed methodology was used to classify LiDAR data and digital aerial images acquired for an urban environment to detect four classes, including trees, buildings, roads, and natural grounds. In this regard, 16 potentially primary features were produced for classification using the LiDAR data and the digital aerial images. The training and checking inputs of the proposed ANFIS were collected from the generated features for further training and evaluating processes. Also, the fuzzy c-mean (FCM) clustering algorithm was used to initialize the fuzzy inference system of the proposed ANFIS-based classification method. By considering all states of fuzzy rules for each training input, the fuzzy rule with the maximum firing value was selected. Accordingly, these fuzzy rules were used as the inputs of rough set theory. Accordingly, the optimum features were acquired by the basic minimal covering (BMC) algorithm as the rule induction method. To validate our proposed methodology, the procedure of classification was repeated by the achieved optimum features. The results showed that the classification using the optimum features has reached better overall accuracy (OA) than those achieved by using the 16 potentially primary features. Also, comparing the results of our proposed methodology with the other well-known GA-based feature selection methods indicated the significance of the proposed FRS method to select optimum features with high accuracy in a short running time.
کلید واژگان :Fuzzy rough set, Adaptive neuro-fuzzy inference system, Airborne laser scanning, Classification, Rule extraction, optimum features.
ارزش ریالی : 600000 ریال
با پرداخت الکترونیک