Study On The Detection Technology For Regional Variation In Remote Sensing Image

Technology for detecting regional variation in remote sensing image has been played an increasingly important rolein civil economy and national defense during past decade.In this thesis several innovative approaches related to detect areal variations in remote sensing image are studied and analyzed, which include1) remote sensing image denoising;2) remote sensing image segmentation; and3) remote sensing image classification.Clouds noise often exitsin the process ofremote sensing imagebecause of the weather conditions.Theregional cloudsin remote sensing image will seriously affect the interpretation and analysisof remote sensing image. By examining the frequency distribution information of the image, remote sensing imagecloud removal method based on HSI color space and wavelet transformation is developed in this dissertation.On the premise ofreserving useful information of remote sensing imagein a large degree,this algorithm can dislodge the cloud noise in remote sensing images well.Remote sensing image segmentation is one of the basic conditionsforthe area change detection. The stand or fall of the segmentation qualitydetermines the success or failure of remote sensing image change detection.Thisdissertation is based on statistics theory and makes use of statistical region merging algorithm to carry on multi-scale segmentation for color remote sensing images.As to theovermerging problem in the process of segmentation, thedissertationcombined with the LBP features and edge features of the segmention regions and put forward a region merging algorithm based on threshold.A good image classification and recognition technology can lead to a good regional change detection result.The words assignment in the traditional Bag-of-Visual-words (BOVW) is studied. A robust soft assignment method is used by studing the words assigenmnt of the hard assignment. The method can improve the classification rate of remote sensing images compared with the traditional approach.It isquite time consuming task for the compute by using the traditional algorithm. By studing the traditional BOVW algorithm,the Fast BOVW algorithmis developed in this dissertation. It can improve the computation speed greatly by improving the approach of visual words vocabulary construction. The experiment carried on the actual database shows that the Fast BOVW algorithmhas fast computation speedabout L (20～30times for our experiments) times with the high degree of precision the same with thetraditional algorithm.The result in this paper has some academic value and significance to improve the ability of handling regional variation in remote sensing image.