SAR image change detection method based on multi-grain cascade forest model

Abstract

The invention discloses a SAR image change detection method based on a multi-granularity cascaded forest model, which mainly solves the problem of strong dependence of the traditional method on the difference map, and its realization scheme is: given two registered multi-temporal SAR images I 1 and I 2 ; to I 1 and I 2 Generating Difference Map I with Log Ratio Operator d ; to I d Extract Haar-like features and input them into the trained multi-granularity cascade forest model to generate two probability maps I 1 and I 0 And with the difference map I d Constitute new features; use the new features to retrain the model to obtain new prediction results. Compare the Kappa coefficient with the previous Kappa coefficient, and select the prediction result with a higher coefficient as the final change detection result. The invention can effectively suppress the influence of the difference map on the final result, improve the accuracy of change detection, and can be used for environment detection and disaster detection.

Description

Technical field [0001] The present invention belongs to the field of image processing, and more particularly to a SAR image change detection method, which can be used in environmental detection, agricultural surveys, natural disaster testing, forest resource monitoring, etc. Background technique [0002] Synthetic aperture radar SAR has high resolution, all day, the advantage of all-weather work is widely used in civilian and military fields. At present, the harsh changes in the natural environment, the rapid development of the city, so that the image change detection technology rapidly rises. [0003] The change detection is used to detect the change in the same location in a period of time, and the traditional change detection method uses Bruzzone et al. In 2002, the classic three-step process paradigm proposed in 2002: 1) Pretreatment; 2) Generate a difference map; 3 ) Analysis diagram. Since the SAR image is influenced by the coherent spots, the detection result of this conve...

AI Technical Summary

Problems solved by technology

[0005] Most of the above methods for change detection using deep learning are based on deep neural networks. Since deep neural networks require a large number of training samples, the number of labeled samples available for training is very limited in practice, which will seriously affect the performance of this method. ;Secondly, there are many hyperparameters in the deep neural network, and it is difficult to adjust the parameters, resulting in low efficiency

Images