Unmanned aerial vehicle railway fastener image shadow removal method and system based on NonshadowGAN

Abstract

The invention provides an unmanned aerial vehicle railway fastener image shadow removal method and system based on a NonshadowGAN. According to the invention, based on the railway fastener image obtained by the unmanned aerial vehicle, the fastener image is firstly preprocessed to realize data augmentation, and then shadow removal of the fastener part is realized by using the NonshadowGAN network. According to the method, when an unmanned aerial vehicle railway fastener image is input into a shadow removal network, a loss function is divided into three parts, adversarial loss is applied to shadow feature learning through a generative adversarial network, shadow and non-shadow feature learning and recognition tasks are completed, available space of a mapping function is reduced through cyclic consistency loss, and the shadow removal efficiency is improved. And further constraining a network structure through identity verification loss, identifying a shadow generation result, and adjusting a parameter model. Therefore, the method can achieve the model training based on the non-paired shadow fastener data, and accurately achieves the removal of the shadow part in the unmanned aerial vehicle fastener data.

Description

technical field [0001] The present application relates to the technical field of high-speed railway detection, in particular to a method and system for removing shadows from UAV railway fastener images based on NonshadowGAN. Background technique [0002] High-speed railway needs to ensure its safe and efficient operation. Existing detection methods, such as manual detection, detection vehicles, etc., have problems such as low detection efficiency, restrictions on repairing skylights, and detection dead angles. In recent years, due to the characteristics of high efficiency, flexibility and low cost of UAV inspection, UAV patrol detection has achieved good results in many fields. Defect detection and monitoring of railway infrastructure based on UAV image acquisition is a new and effective method. However, in practice, in order to avoid interfering with the normal operation of the train, drones are not allowed to shoot above the railway line. However, the UAV image data acq...

AI Technical Summary

Problems solved by technology

Moreover, in the training image, we can only generate shadow pairing data in a small range, and cannot introduce data into large scenes or scenes that are not controlled by shadows

Due to the above limitations of the training data, the railway fastener data collected by the drone belongs to the scene that cannot be manually matched in the prior art, and we cannot provide paired shadow fastener data

Images