SAR Image Change Detection Based On Deep Forest

Synthetic Aperture Radar(SAR)image change detection is that the analysis of SAR images in the same region but at different times to obtain the change information of multi-temporal images.In recent years,remote sensing data has become the main data source for change detection which provides a more convenient way for change detection,and change detection technology for remote sensing image has been widely used in various aspects.However,due to the scattering of SAR images,there is a large amount of speckle noise in the image,which seriously affects the quality of the image.Therefore,it is necessary to extract important feature information in the image and effectively suppress the noise,thereby,the accuracy of image detection can be improved.At the same time,reducing the complexity of the algorithm and improving the anti-noise performance of the algorithm are one of the problems that need to be solved.Therefore,in order to improve the performance of change detection,the main contents are as follows: 1)The SAR image change detection method based on deep forest model is implemented.It mainly deals with the problems of more speckle noise and less SAR image data in SAR images.In the pre-classification process,the non-local mean algorithm is used to process the difference map according to the neighborhood information of the image,which can maintain the balance between the denoising of the difference map and the detail retention of the difference map,thereby improving the detection result of the pre-classification.Because the deep forest model has fewer parameters,and the parameters are robust and suitable for small data sets,the deep forest model can improve the accuracy of detection and simplify the training process.Experiments show that the implemented algorithm can fully learn image features and achieve effective change detection.2)The SAR image change detection method based on mixed feature representation learning is proposed.It because that the characteristics of the difference image block will affect the training effect of the model.Firstly,three different images with different characteristics are formed by using wavelet fusion method,sparse self-encoding feature extraction method and non-local mean method.Then,the characteristics of three different images are learned based on the deep forest model.The results of the data set training of different features are combined to obtain the final test results.The method is based on the integrated learning strategy combined with multi-features to achieve change detection.The experimental results show that the learning of mixed features can improve the accuracy of detection.3)The SAR image change detection method based on multi-scale fusion and edge information is proposed.It is mainly for the problem that the size of the image block input has an effect on the result and the edge portion is difficult to detect.The method adopts multiscale model and incorporates gradient information to improve the image size factor and improve the accuracy of edge detection.This method constructs a multi-scale deep forest model,which can learn multi-scale image features and can reduce the influence of local information of the image on the classification results.In addition,through experimental analysis,most of the error pixels detected by image change exist in the edge part of the image.Therefore,the method combines the image class probability map with the image edge information.The experimental results show that the method can improve the accuracy of the change detection result.