Aeromagnetic Image Information Extraction Method Based On Level Set Image Segmentation

There are two methods for the analysis of the information contained in the aerial iamge, one is artificial interpretation, and another one is the computer interpretation. The advantage of artificial interpretation lies in the subjective experience, but sometimes it will make mistakes and the advantage of computer interpretation lies in the objective analysis, and can be faithfully execute the analysis of intension and not be affectd by others. With the development of computer and information technology, computer interpretation will be more and more attention, and play an important role in the field of aerial image analysis. From the present situation, the main problems of computer interpretation are concentrated in the field of image sementation and feature extraction methods are not mature enough.In this paper, to make the model can handle the image with noise, the principles of image denoising wavelet, wavelet function and wavelet transform decomposition level for denoising image have been studied. By comparing the classical median filtering with Wiener filter and mean filter denoising experimental. Results show that after wavelet denoised image fusion Chan. Vese model image segmentation has a best result. Aeromagnetic image after wavelet transform and Chan. Vese model after merging, the resulting image is divided not only removes the edge segmentation error, while dividing the contours become more smooth. In this paper, made in-depth analysis based on the traditional curve evolution theory. At the same time the traditional correlation energy level set evolution model has been inferenced. To overcome the shortcomings of Mumford-Shah model and Chan-Vese model, Local Chan-Vese local information(LCV) model are presented. The results showed that the kind parting model can approve the target of extremal regions segmented image, it can be the foundation of aeromagnetic image identification. And Three encoding methods have been proposed based on the image of the cross-boundary contour, columns and oblique. Various level set models are mainly used to extract image boundary contour of the curve and then encoded getting the contour boundary feature attributes without image. Because different images have different boundary contour, there are big differences between the same coding strands by different encoding. By analyzing different encoding methods to, the white value obtained by comparing the coding section transverse width, a smaller number. So for the single-phase image feature attributes characterizing the most obvious. It has high anti-interference performance and is suitable for the identification process of the single-phase images. However, the value segment of white columns and diagonal coding get narrower ebcoding, a larger number of distributions and are more suited to the multiphase image identification process. In addition, on the basis of this kind of coding strand established on BP neural network model to be used for image recognition. In the experiments of recognition of mechanical artifacts, this kind of identification with high accuracy, while based on the image of the boundary contour coding strand, samples of the same type can be identified requiring only one image of the sample. So some problems can be solved,which are slow calculating speed, large sample size and does not recognize the image contains strong noise and other issues by this kind of model. Finally, this kind of hybrid model is used in identifying the extreme area aeromagnetic image of Northeast region of China. It is proved that this method can improve the efficiency and quality,and has practical significance to the development and research of computer interpretation of aerial magnetic image.