On-Orbit Change Detection Method Research For SAR Image

Change detection of the image is aimed at finding the change information of the regio n in the same area at different times,change detection of SAR image is an important application of synthetic aperture radar.Traditional change detection method need to transmission massive data to the ground firstly,the large amount of redundant data and the limited transmission bandwidth of satellites have seriously affected the timeliness of demand.The method which training network on the ground and detect changed areas on orbit can not meet the needs of many types of scenes because of its fixed netwo rk parameters.Based on the above problems and combines the advantages of high speed and efficient classification of ELM,this paper are proposes some methods to automatically extract the training samples from the image which will be detected,and realized on-orbit SAR image change detection verification system.Specifically,we will study from the following aspects:(1)An on-orbit change detection method based on fuzzy C-means clustering(FCM)algorithm and extreme learning machine(ELM)is proposed.In this method,labeled samples are extracted by the sample selection criteria from different image,the neighborhood feature of the partially labeled sample is extracted as the training data of the ELM,and finally the changed areas are detected by trained ELM.In this method,the principal components of the difference image are extracted by PCA,which reduces the interference of the noise pixels to the detection results.The labeled samples extracted with the FCM and the sample selection criteria,which avoid the requirement for artificial samples and optimized the ELM classification results.The experimental results show that the FCM and ELM methods proposed in this paper are superior to other contrast methods in denoising performance and total error rate.(2)Combining the texture information of SAR images,we proposed an on-orbit change detection method based on distance AP(DAP)clustering and Affinity Regularized Extreme Learning Machine(ARELM).The method divides the difference image by the SLIC super pixel segmentation method,and use super-pixel substitutes the pixel as the sample of the DAP.When the labeled samples are extracted from the samples,the distance dimension between the samples is added to the decision process of the AP algorithm.The K-means clustering algorithm is used to automatically classify the change samples,the unmodified samples and the unknown sample.The ARELM is used to detect the SAR region of the SAR image.The SLIC super-pixel segmentation method is used to preserve the local texture features of the image while greatly reducing the number of samples.The improved clustering results of the improved DAP algorithm provide more generalized training data.The use of the local consistency between the pixels to construct the graph regular term makes the classification results stable and accurate.The experimental results show that the detection accuracy and Kappa coefficient of this method are higher than other results.(3)Based on the gray scale distribution of the image and the characteristics of the pixel classification probability in the process of change detection,a sample selection strategy based on similarity is proposed.Using the human visual characteristics,the local experience field is applied to ELM,and a semi-supervised ELM-LRF is proposed.In the process of applying the optical image to the change detection,a heterologous detection strategy is proposed.Using the sample selection strategy based on similarity to automatically select the required sample samples,training semi-supervised ELM-LRF network,and finally through the heterologous detection strategy to give the final test results.Due to the excellent strategy and network characteristics in the method,the correctness of the method in each data set has been further improved.(4)In this paper,the change detection and verification system of on-orbit SAR image is realized,and the interactive interface of the change detection verification system is designed.The code of the on-orbit SAR image change detection method based on DAP and ARELM is compiled,and the change detection system is tested finally.The actual test results prove the reliability and practicability of the change detection and verifica tion system.