Unmanned aerial vehicle image change detection method based on semantic segmentation and twin neural network

Abstract

The invention belongs to the technical field of multi-temporal unmanned aerial vehicle image change detection, and particularly relates to an unmanned aerial vehicle image change detection method based on semantic segmentation and a twin neural network. The method comprises the following steps: S1, expanding a data set and dividing the data set; S2, establishing a deep neural network model based on combination of a semantic segmentation framework DeeplabV3 and a twin network; S3, training a DeeplabV3-based twin neural network model by using the training data set; and S4, based on the test dataset and the trained model, verifying a training result. According to the method, the semantic segmentation thought is combined, the weight sharing characteristic of the twinning network is utilized,features with realistic meanings are extracted, the semantic relation between pixels and the multi-scale problem of a change area are considered, the problems of noise sensitivity, low change detection precision and the like are solved, and the quality and robustness of a difference graph are improved.

Description

technical field [0001] The invention belongs to the technical field of multi-temporal UAV image change detection, and more particularly relates to a UAV image change detection method based on semantic segmentation and twin neural network. Background technique [0002] In recent years, with the vigorous development of UAV technology, the combination of UAV's own characteristics and aerial photography has become a new development direction. UAV remote sensing technology has the advantages of real-time, high resolution, high cost performance, and strong flexibility, and is used in ecological environmental protection, land use investigation, and river detection. [0003] UAV multi-phase image change detection (Change Detection), in essence, is to detect the significant change area of ​​two images at the same location and different phases, but due to the influence of lighting, weather, camera factors, etc., it has become a An important factor that restricts the play of change de...

AI Technical Summary

Problems solved by technology

[0006] At present, many scholars are studying change detection technology. The paper "A Neighborhood-Based RatioApproach for Change Detection in SAR Images" proposes a neighborhood-based ratio operator, which combines grayscale information with spatial information of adjacent pixels. To generate a difference map, but the feature robustness is not high; the paper "Unsupervised Change Detection in Satellite Images Using Principal Component Analysis and k-Means Clustering" proposes an unsupervised change detection that combines principal component analysis with K-means clustering, However, the feature expression of manual features is not high; the paper "Change Detection Based on DeepFeatures and Low Rank" proposes a multi-scale segmentation and low-rank decomposition method to solve the change detection problem from the perspective of visual saliency, but it loses to a certain extent detection accuracy

The above methods all implement change detection from the perspective of pixels, without taking into account other geometric, semantic information and multi-scale change areas, and do not solve the problem of image change detection being sensitive to noise

Images