Abstract:

A change detection model based on neighborhood correlation images(NCIs)and decision tree classification using remote sensing data was proposed, and then applied to detect forest landscape change information induced by forest disturbance.Longxi-Hongkou nature reserve which was seriously damaged in 5.12 Wenchuan Earthquake was selected as study area to verify the model, and various neighborhood configuration of correlation images were explored using bi-temporal Landsat5 TM images. Change detection maps were generated by using a machine learning decision tree(C5.0). The results shows that the accuracy of change detection results using NCIs is higher than that of result without NCI. Result with 5×5 window size is of highest accuracy among the different NCIs, and general accuracy and Kappa coefficient is 82.33% and 0.808 5 respectively.

Key words: change detection, neighborhood correlation image, decision tree