Image super-resolution reconstruction method for electric power grid inspection robot
A technology for super-resolution reconstruction and power grid inspection, applied in image data processing, graphic image conversion, instruments, etc., can solve the problems of small occupied area, large occupied area, low resolution, etc. The effect of improving accuracy and good extraction effect
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0058] Before model training, you first need to obtain appropriate training and testing datasets. The data set used in this embodiment is composed of a combination of public data sets and self-built data sets. The training dataset contains a total of 1,300 images, of which 1,000 are from the DIV2K dataset, and the other 300 are high-definition images of power components taken at the inspection site. There are two test data sets, one is the public data set BSD100, which contains 100 images, and the other is the images of power facility components taken at the inspection site, which contains 50 images, which constitute a self-built test set. All images are bicubic downsampled to obtain a low-resolution image corresponding to the high-resolution image, which is used to simulate a natural image1.
[0059] The structure of the image super-resolution reconstruction network in this embodiment is as follows figure 1 As shown, the pre-convolutional layer 2 is an ordinary convolutiona...
Embodiment 2
[0066] In this embodiment, an ablation experiment is performed on the basis of Embodiment 1, and the modulation feature map SK in Embodiment 1 is removed from the CS attention mechanism 33 and input to the IK attention mechanism 34, and other parts of the network remain unchanged. The modified two attention The structure of the force mechanism is as Image 6 shown. Using the same conditional training and testing model as in Example 1, the comparison results are as follows:
[0067]
[0068] From the results in the above table, it can be seen that inputting the modulation feature map SK from the CS attention mechanism 33 to the IK attention mechanism 34 can significantly improve the performance of the model and help improve the super-resolution reconstruction effect.
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


