Deep learning-based TFDS non-fault image detection method

Abstract

The invention discloses a deep learning-based TFDS non-fault image detection method. The method comprises the steps of building a network model capable of synchronously detecting multiple regions and multiple types in combination with deep learning image classification and target detection technologies based on a deep learning theory; performing locating detection on regions of non-fault key parts in railway TFDS images by the model; judging the images, in which the regions of the non-fault key parts are not detected, as suspected fault images; and observing and detecting a small amount of the suspected fault images by human eyes of indoor train inspection personnel. According to the TFDS non-fault image detection method provided by the invention, a large amount of the non-fault images are filtered out and only a small amount of the suspected fault images are observed and judged by the human eyes by virtue of the deep learning technology; and the method can greatly reduce the workload and working intensity of the indoor train inspection personnel.

Description

technical field [0001] The invention relates to the field of railway detection, in particular to a TFDS non-fault image detection method based on deep learning. Background technique [0002] The rapid development of my country's railways has become an important means of transportation for people to travel and transport goods. While bringing convenience to people, people have higher and higher requirements for railway safety performance. Any small fault can cause major accidents. At present, the fault detection of freight trains and railways is based on the TFDS system (freight train fault image dynamic detection system), and the train inspectors in the monitoring room observe the collected images through the image reproduction terminal software, relying on human eyes to observe the images, and a large number of trains running every day need A large number of indoor train inspectors come to observe the image, and it is a huge workload for each indoor train inspector. [0003...

AI Technical Summary

Problems solved by technology

These methods have high requirements on the scene of the fault and the form of the fault, and the types of faults are often diversified, and the forms of faults that may occur in a certain area are also diverse, so these automatic fault identification algorithms are not efficient. It is easy to miss detection, and the practicability is not high. Relying on these methods to automatically identify the fault of the train on the spot will have a great risk

Images