Unlock instant, AI-driven research and patent intelligence for your innovation.

Railway image style conversion data amplification method and system

A style conversion and image technology, applied in the field of rail transit security identification, can solve the problems of poor algorithm stability, easy to be affected by the environment, and difficult to collect high-speed railway intrusion samples, etc., to achieve the effect of image pixel loss and stable quality

Pending Publication Date: 2022-07-29
BEIJING JIAOTONG UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The former includes frame difference method, optical flow method and background difference method, etc. Its detection speed is fast, but it is easily affected by the environment and the algorithm stability is poor; the latter mainly includes various algorithms of convolutional neural network, which can be divided according to the function of the algorithm. It is a semantic segmentation algorithm and a target detection algorithm; according to the detection idea, it can be divided into a target detection algorithm based on candidate regions (Two-stage) and a target detection algorithm based on regression (One-stage). The former mainly includes R-CNN series algorithms, etc. The latter mainly includes YOLO, SSD and other algorithms, but no matter what kind of algorithm it is, it has a high dependence on the number of data sets. The accuracy of various deep learning algorithms is highly dependent on the labeling accuracy and richness of training samples.
In the high-speed railway scenario, due to the characteristics of high-speed railway operation during the day and maintenance at night, it is difficult to collect a large number of high-speed railway intrusion samples for training. In addition, there are various differences in high-speed railway samples in different regions. The image quality (2k, 1080p, 720p) of the intrusion samples varies from place to place, so it is difficult to further improve the accuracy of the algorithm

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Railway image style conversion data amplification method and system
  • Railway image style conversion data amplification method and system
  • Railway image style conversion data amplification method and system

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] The present embodiment 1 provides a railway image style conversion data augmentation system, and the system includes:

[0041] The acquisition module is used to acquire the relevant images of the railway environment to be processed;

[0042] The processing module is used for using the improved generative adversarial network to process the relevant images of the railway environment to be processed to obtain the images after seasonal conversion; wherein,

[0043] The improved generative adversarial network is: the image is convolved and normalized in the network, and then activated by an activation function. After two consecutive downsampling, the output image is subjected to multiple residual blocks, and the output results are added in order. Channel attention mechanism and spatial attention mechanism, in which two convolutions are performed in a residual block; after adjustment by the attention mechanism, the image and retrograde are continuously upsampled twice; finall...

Embodiment 2

[0057] In this embodiment 2, a data augmentation method for railway image style conversion is provided, which solves the problem of obtaining a data set cost-effectively and efficiently when the training sample data is insufficient by using a cycle gan network model with an attention mechanism added.

[0058] A data augmentation method for railway image style transfer based on attention mechanism, including the following steps:

[0059] Step 1, obtain the railway-related images, and start the improved generative adversarial network;

[0060] The step 1 specifically includes the following steps:

[0061] Step 1.1, perform size processing on the input railway perimeter image, first scale it to 560*560 size, then crop it to 512*512 size image and input it into the network;

[0062] Step 1.2, start generators A, B, and input the scaled railway image into the network.

[0063] Step 2, the image undergoes 7*7 convolution, the HW is normalized, and then activated by the activation ...

Embodiment 3

[0081] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor, the processor and the memory communicate with each other, the memory stores program instructions that can be executed by the processor, and the processor invokes the The above-mentioned program instruction executes the railway image style conversion data augmentation method, and the method includes the following process steps:

[0082] Obtain the relevant images of the railway environment to be processed;

[0083] Using the improved generative adversarial network to process the railway environment-related images to be processed, the seasonally converted images are obtained; among them,

[0084] The improved generative adversarial network is: the image is convolved and normalized in the network, and then activated by an activation function. After two consecutive downsampling, the output image is subjected to multiple residual blocks, and the output results are added i...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides a railway image style conversion data amplification method and system, and belongs to the technical field of rail traffic safety identification. And processing the railway environment related image to be processed by using the improved generative adversarial network to obtain an image after season conversion. According to the method, a result of superposition calculation is continuously trained by means of a network public data set submer2winteryomite and a railway data set possessed by a laboratory, a generated picture can be identified by a corresponding target style identifier, an output picture has a good style conversion characteristic, and compared with an initial network, the method has the advantages that the efficiency is high, and the cost is low. According to the method, the problem of image pixel loss generated after low-resolution pictures are input can be well solved, the quality of image generation is more stable, and the method is more suitable for data amplification.

Description

technical field [0001] The invention relates to the technical field of rail traffic safety identification, in particular to a railway image style conversion data augmentation method and system. Background technique [0002] At present, most of the traditional railway perimeter environmental safety protection adopts the method of human inspection. On the one hand, this kind of method is not only difficult to detect the existing safety hazards in time, but also may pose a huge threat to the life safety of the inspectors under certain conditions; On the one hand, because the high-speed railway network is widely distributed, such a large inspection will inevitably waste huge manpower and material resources. [0003] For this reason, the automated high-speed railway foreign body intrusion detection came into being. At present, the detection methods are mainly divided into two categories: contact detection methods and non-contact detection methods. Contact detection methods such ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06V10/774G06V40/10G06V10/82G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/048G06N3/045G06F18/214
Inventor 谢征宇陈俊明贾利民秦勇
Owner BEIJING JIAOTONG UNIV