Transmission line aerial image data cleaning method based on content awareness

Abstract

The invention discloses a content awareness-based power transmission line aerial image data cleaning method, which comprises the following steps of: firstly, constructing a self-classification model, and cleaning an image which does not contain an insulator part in an aerial image; training a model for extracting image content by using the residual data set; the method comprises the following steps: acquiring local and global features of an image, establishing a quality perception rule according to different influence degrees of local and global contents on image quality, mapping an image quality score function, and finally screening the image according to image quality scores. According to the invention, mass aerial image data can be effectively cleaned, so that clearer and higher-quality images can be obtained.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a method for cleaning aerial image data of transmission lines. Background technique [0002] Electric power departments need to conduct regular power inspections, and drones have become a trend to replace manual inspections. The information extracted from aerial images is used to locate the faults of insulators of transmission line components. However, drones will collect a large number of invalid images, and some of the images will be distorted during the transmission process due to hardware equipment; too much invalid data, on the one hand, will lead to a surge in the amount of manual detection tools, reduced accuracy, and reduced timeliness; On the other hand, it is not conducive to building a better and more practical insulator defect detection model. Therefore, cleaning the massive aerial images of transmission lines and screening out high-quality image...

AI Technical Summary

Problems solved by technology

The problem is that the traditional algorithm needs to convert the image into a fixed-length feature vector, which will affect the quality of the image; in addition, the scalability is not very good, and different application scenarios require different algorithms

The CG-DIQA method proposed in 2018: Convert the input image to a grayscale image, scale it to a fixed size, and estimate the quality score of the document image after calculating the standard deviation. This method cannot predict the quality of occluded images.

The problem is that CNN requires that the input image size is fixed, some need to be cropped to a fixed shape, some need to be resized, and cannot be well adapted to various real images, because real aerial images have different lengths and widths Proportion

After the vision transformer was proposed, the Norwegian Research Center applied the transformer to image quality evaluation for the first time, overcoming the CNN model’s image resize operation, and being able to adapt to image resolution; however, it was also experimented based on KonIQ-10K, although the quality has been improved The prediction effect, but the local distortion of the image is not considered too much, and the distortion type of the database is not complicated, which is lacking compared with the distortion in the real environment

Images