Mine mobile inspection image reconstruction method based on edge correction

Abstract

The invention discloses a mine mobile inspection image reconstruction method based on edge correction, and the method comprises the following steps: 1, carrying out the image preprocessing: carrying out the downsampling and image segmentation of different degrees on an input image; 2, extracting and representing image features; respectively using an improved activation function behind each convolution layer; 3, performing image reconstruction, adopting the method to reconstruct the Y channel; and 4, correcting edge error area information. In the preprocessing stage, the images of the training set are amplified at different scales to facilitate later cross training, so that the network model is suitable for reconstruction processes of different scaling multiples of 2, 3, 4 and the like. Network levels are properly deepened, and richer image feature information is extracted. An edge correction coefficient is extracted, edge information correction is performed on the reconstructed HR image, and the problem that edge details are fuzzy is solved. An improved activation function is introduced, the nonlinear expression capability is improved, and more nonlinear region characteristics are activated.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image reconstruction method for mine mobile inspection based on edge correction. Background technique [0002] In the current mine system, there are many coal conveyor belt conveyors and transported coal streams, which makes it difficult to manually detect whether the belt transport is safe or not. Therefore, mobile inspection methods are often used in daily production to automatically and efficiently detect the status of coal stream transportation. The detection process Due to the problem of the camera environment, the captured video image will be unclear and distorted, which will affect the detection effect. Therefore, it is necessary to perform super-resolution reconstruction on the monitoring image of the mobile inspection. [0003] Through the mobile inspection image, the transportation of coal flow and the safety status of underground workers can be checked in r...

AI Technical Summary

Problems solved by technology

[0006] The interpolation-based method is difficult to reproduce the detailed information such as the texture of the image, and the generated image is relatively blurred; the reconstruction-based method usually requires computationally complex image registration and fusion stages, and its accuracy directly affects the quality of the result; the learning-based method Among them, SC needs a large number of high and low resolution image blocks to train high and low resolution dictionaries, which takes a long time; SRCNN has too few convolutional layers, small receptive field, limited image features, and can only be applied to one kind of enlargement Network training under multiples; Kim et al. chose a deeper convolutional neural network, the overall structure is more complex, the training time is longer, and the amount of calculation is larger; FSRCNN reduces the number of parameters and reduces the amount of calculation while also reducing the reconstruction effect; Xiao Jinsheng's method uses the pooling layer to reduce the dimensionality, which will lead to the loss of a lot of detailed information of the image, which will affect the accuracy of super-resolution reconstruction.

Images