Polarized SAR image classification method based on CNN and RFC ensemble learning

Abstract

The invention discloses a polarized SAR image classification method based on CNN and RFC ensemble learning, and mainly solves the problem that an existing polarized SAR image classification method isnot high in classification precision. The method comprises the following steps of: extracting T matrix and Cloud decomposition characteristics as original characteristics according to a filtered polarimetric SAR image; randomly selecting 1% marked samples as training samples, marking the samples as TrainPixel, and training the RFC model by using the TrainPixel; extracting a 21 * 21 neighborhood block of each pixel point in the polarimetric SAR image, taking the 21 * 21 neighborhood block as an input feature of the CNN, and recording the input feature as F2; selecting a neighborhood block corresponding to TrainPixel from F2 as a training sample of the CNN, and training a CNN model; and obtaining a boundary region of the polarimetric SAR image by using information entropy based on a classification result of the CNN model for the whole image, and classifying the boundary region and a non-boundary region by using RFC and CNN respectively. According to the method, a CNN and RFC integrated learning method is used, the advantages of the two methods are comprehensively utilized, and a good classification result can be obtained in a boundary region and a non-boundary region of the polarimetric SAR image.

Description

technical field [0001] The invention belongs to the technical field of image processing, and mainly relates to a polarization SAR image classification method, specifically a polarization SAR image classification method based on an integrated learning method based on CNN and RFC, which can be used to classify polarization SAR images object classification and object recognition. Background technique [0002] Polarimetric SAR image classification is a very important application in the field of remote sensing processing. Polarization SAR is not affected by time and weather, and contains rich polarization information, so polarimetric SAR has been successfully applied in many fields such as agriculture, military affairs, geological exploration, urban planning and ocean monitoring. Over the past few years, a large amount of polarimetric SAR data has been put into application. Therefore, polarization SAR image classification has attracted the attention of many scholars, and a larg...

AI Technical Summary

Problems solved by technology

The classification method based on scattering characteristics is usually based on physical meaning, and sometimes it is combined with other characteristics in order to obtain better classification results. However, such algorithms require rigorous analysis and derivation, which raises the processing threshold of polarimetric SAR data. The further development of such methods is limited; classification methods based on statistical characteristics are usually based on Wishart distribution, but it takes a long time to calculate Wishart, and only based on one distribution is not conducive to obtaining good classification results, so it is not conducive to use such Algorithms to deal with polarization SAR image classification problems; classification algorithms based on machine learning often use only one method to deal with polarization SAR image classification problems, due to the complex characteristics of polarization SAR data, it is difficult to use only one machine learning method Obtain ideal polarimetric SAR image classification results. For example, CNN has achieved good classification results in polarimetric SAR image classification, but since CNN needs to use the neighborhood of pixels as the model input to obtain the classification results of the pixel points, CNN The classification results in the image boundary area are not satisfactory. RFC also obtains good classification results in polarimetric SAR image classification, but RFC does not obtain the spatial information of the image, so the overall classification result of RFC is not as good as CNN, but in the image The boundary area is better than the classification result of CNN

Images