Research On Lithologic Classification Of Remote Sensing Image Based On Object Oriented

In recent years, with the rapid development of remote sensing technology, remote sensing application areas are more open, remote sensing technology as an advanced space exploration technology has been more attention. Classification is the key technology in remote sensing application, traditional classification is based on pixel as the basic unit, according to the principle of foreign body in different spectrum, in the feature space based on its statistical characteristics to achieve classification. Due to noise and the neglect of the local information, classification results appeared serious phenomenon of salt and pepper, expressed spatial distribution weakly, unable to make full use of the information contained in the remote sensing data. And object oriented classification method formed by segmentation of image objects as the basic unit of classification, image objects is by the merger of the local pixel, and the impact of image objects can be ignored, so the object oriented classification method can avoid the deficiency based on the two aspects, such as element classification method, thus avoid the phenomenon of salt and pepper.In this paper, the object oriented method is used to classify lithology. The homogeneous image objects are generated by segmentation technology, then extracted feature information of object, and classification by using the method of fuzzy classification. Classification experiments were performed using Landsat8 images in Guyang area, Baotou city. Some processing needs to be carried out before the classification. Firstly, use Gram-Schmidt transformation fusion method for Panchromatic and multispectral bands. Then calculate the optimal index of the combined band, select the band 651 for false color synthesis. Finally, principal component transformation on the fused image and extract texture information on the first principal component. After repeated segmentation experiments, the final selection of the scale is 120, 90, 60, the three levels to establish a classification network, based on the spectral and textural features of the object, the classification is realized by using the nearest neighbor classification method. Finally, the classification result is evaluated by the geological map. The results show that the overall accuracy of object-oriented classification is 84%, which is improved by 11% compared with the maximum likelihood method, the Kappa coefficient of object-oriented classification is 0.82, which is improved by 0.12 compared with the maximum likelihood method. It can be seen that the object-oriented classification method has certain application value in lithology classification.