Group building damage remote sensing classification method based on improved CNN

Abstract

The invention discloses a group building damage remote sensing classification method based on an improved CNN (Convolutional Neural Network). The group building damage remote sensing classification method comprises the following steps: A, image preprocessing: performing geometric correction, orthographic correction and image registration on a remote sensing image; b, image segmentation: segmentingthe remote sensing image according to the block vector data; c, image block cutting: sequentially cutting image blocks with specified sizes in the minimum enclosing rectangle of each segmented block;d, image block screening: if the overwrapping proportion of the image blocks and the located blocks is greater than 50%, the image blocks meet the requirements, and a certain number of effective image blocks are screened from each block; e, image block prediction: inputting each effective image block into the trained basic CNN Inception V3, and predicting the probability that the effective imageblock belongs to each damage category; and F, block prediction: integrating the category probabilities of all effective image blocks in each block to obtain the damage category probability of each block. The method is easy to implement and convenient to operate, and a more accurate building damage remote sensing classification result is obtained.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and more specifically relates to an improved CNN-based remote sensing classification method for group building damage, which is suitable for optical remote sensing images with a resolution of 0.5m-2m. Background technique [0002] In the rescue and recovery phase after an earthquake, because the damaged building may be a representative of the location of the trapped people, quickly realizing the classification of building damage and obtaining its classification map is an urgent need for post-earthquake golden time rescue, and it is also a post-disaster A key basis for disaster loss assessment and post-disaster reconstruction. Although the accuracy and confidence of building damage information obtained by traditional manual field survey methods are high, there are many shortcomings such as heavy workload, high cost, and low efficiency for large-scale surveys, and the inform...

AI Technical Summary

Problems solved by technology

First, CNN is used to automatically select the optimal feature to solve the problem of difficult feature selection in the object-oriented process; then, the block is used as the smallest classification unit to generate a meaningful boundary for the group buildings in the image, effectively replacing the traditional image segmentation , to solve the problem of unsatisfactory general image segmentation; then add Separate and Combination layers to the basic CNN Inception V3 to improve the traditional CNN, so that it can directly process remote sensing images and block vector data, and solve the problem that general CNN cannot directly process block data. The Problem of Predicting the Damage Category of Group Buildings

Images