An Improved Segmentation Method And Application Of UAV High-resolution Remote Sensing Image Based On Mean Shift

With the development of remote sensing technology, such as UAV remote sensing technology, the approach to improving the speed and accuracy of remote sensing images of extract information processing is an important research direction in the field of remote sensing. Remote sensing image segmentation in the process of remote sensing information extraction technology is the most important one, while it is also a difficult research subject of remote sensing image processing. In the image processing, image segmentation method has yielded some achievements, however, in the high-resolution remote sensing image segmentation, segmentation accuracy is not high and inefficient. To solve the above problems, this paper intends to use a highly efficient and robust Mean Shift algorithm for high-resolution remote sensing image segmentation in order to improve the accuracy and reliability of remote sensing image segmentation.(1) Due to the light weight UAV large dependence on weather conditions, it can be affected by unstable weather aircraft attitude and image distortion of the camera itself. Thus it is necessary to get the original image preprocessed. In this paper, INPHO software adjusts the original images as well as aerial triangulation, generating orthophotos and other treatment, which would provide accurate experimental data for subsequent segmentation of remote sensing images.(2) The boundaries blur and the accuracy of traditional Mean Shift algorithm for remote sensing image segmentation is not high, so this paper proposes a new texture features based on improved Mean Shift LBP remote sensing image segmentation. By using traditional Mean Shift segmentation algorithm, only the images spatial position and color as the feature vectors can be involved in the calculation, which results in limited increase segmentation accuracy. The proposed method in the traditional Mean Shift algorithm with texture forms a "spatial location, color and texture" feature vectors and adaptive clustering. The results show that:compared to the traditional remote sensing image segmentation method, improved Mean Shift segmentation Mean Shift can get better segmentation.(3) Since improved Mean Shift algorithm in remote sensing image segmentation experiences the processing, the regional similarity criteria preclude the use of this article and an area-weighted minimum area parameter for remote sensing image object region merging divided. In order to solve the improved Mean Shift segmentation algorithm compares pieces, segmentation method proposed in this paper can get better segmentation. The results show that:compared with the traditional remote sensing image segmentation method, the results after treatment with EDISON software and eCognition software shows that the improved Mean Shift algorithm can get better segmentation results, which can improve the accuracy of remote sensing image segmentation to some extent.