Research On Multi-scale Feature Extraction And Fusion Method Of Multi-Source Remote Sensing Image

The realization of target location after feature extraction of typical target in remote sensing image is not only a popular research topic within the field of image processing, but also a hot spot in the field of automatic target location. Airport as an important means of transportation and the strategic hub, it’s automatic location has great significance in both military and civilian areas, and it has attracted widespread concern.Feature extraction can get information that reflecting the target characteristics from the large remote sensing image data, so as to get symbols of intuitive significance and consistent with the visual features. It is the core of target location and the precondition of classifier accurately detecting the target. The stand or fall of feature extraction directly affects the accuracy and reliability of target location. The linear feature extraction of runway in airport target location is a very important part. For the problem of the linear feature extraction effect based on Canny operator edge detection is not ideal, this paper designs a method using the multi-scale characteristics of wavelet for edge detection. The edge can be detected by and large under the large scale and refined under the small scale. Then using Hough transform for straight line extraction to select the parallel lines according to the angle, the distance, the slope between two lines and the length of the straight line. This method can better extract the linear feature of airport runway.Multi-source image fusion can make full use of images from different sensors. Extracting feature from these images and making a fusion of them can result a new feature vector that of more valuable information as to achieve more accurate and higher reliability target location. For the unicity of the image source and feature extraction in airport target location, it designs a method based on multi-source remote sensing image fusion for target location. In view of the characteristics about multi-spectral image, panchromatic image and SAR image and the complementarity for the same scene, this method extracts some features from these three kind images. First of all, it extracts the global contour features of the multispectral images and gets the airport potential areas according to the areas of the contours. Secondly, it extracts the local texture, such as textural features, color features, edge line features and back scattering characteristics of potential areas. Finally, all the features above combined into a new feature vector is put to the SVM classifier so as to determine whether the image contains the airport or not. The global features are difficult to describe the details of the image, and the local characteristics can explain the details of the image from the region, the two feature complement each other. The method combines with the characteristics of multi-source image and has a higher accuracy.