Research On Interested Land Feature Change Detection Algorithms Of Polarimetric SAR Image Based On Classification

Synthetic Aperture Radar(SAR) which possesses the characteristic of all-weather, all-time capabilities, has become one of the important means to obtain information in the field of remote sensing. With the development of multi band, multi polarization and full polarization SAR, the scattering information of target detected by radar is becoming more and more plentiful, it makes the scattering mechanism of target could be extracted more effectively and greatly, it also promoted the adoption of SAR in many fields such as target classification, target detection and target recognition. Change detection of Polarization Synthetic Aperture Radar image is an important branch of image processing and plays a significant role in both economic field and civil field, It can be used in many fields like monitoring changes in topography, preventing flood and waterlogging, searching and rescuing after disaster, detecting and evaluating urban change, monitoring the growth of farmland, perceiving battlefield and etc. Most of these applications, However, mainly focus on the changes of specified land cover, for example it's a common practice to detect the changes of water for flood prevention a, detect the changes of building group for assessment of city changes and detect the changes of farmland for monitoring the growth of farmland. As the matter of fact, polarimetric SAR data usually includes many different types of land cover, those non-interest land feature also probably change a lot during the observation time. Thus this paper mainly focus on how to use polarimetric SAR data to achieve change detection of the specified interest land cover in complex sense.Two detection algorithms will be proposed in this paper to achieve change detection of interested class of land-cover in complex sense.(1). An improved Wishart distance algorithm is used to measure the distance between the sample of land cover and two phase images,Then, the ratio of logarithmic transformation is used to obtain difference map. Finally,double threshold segmentation method is apply to extract the change areas and also distinguish two kinds of changes.(2).The h/q decomposition is used to classify each pixel of two phase images respectively, then bayes iterative classification method is adopted to re-classify in original phase images and after that final classification result can obtain. Secondly, both polarization scattering characteristics and echo characteristics are apply to extract the pixels which matching the demand, and then binary images which only contain interested land feature are obtained. Finally, double difference operation is used to obtain two kinds of change areas.The first algorithm base on supervised classification, which needs the training sample of interest land feature. This algorithm has small amount of calculation for no need the initial classification. The second algorithm, which was based on unsupervised classification could overcome the influence on results from the sample training, meanwhile, the threshold segmentation algorithm is replaced by binarization extraction in the interest land feature of two phase images, those improvements greatly reduces the influences of threshold setting. Multi-group of radar data collected by Radarsat-2 are used in this paper, the experimental results shows that those two algorithm gives a good performance to achieve the change of the specific land cover type in complex scenes.