A Lightweight Denoising Method Based on Noise2Void for X-ray Pseudo-color Images in X-ray Security Inspection

For public security and crime prevention, dualenergy X-ray detection technology has been widely used in the security inspection. Due to the special working environment of the security checker, the generated X-ray pseudo-color images contain a lot of noise. While the image denoising methods based on deep learning have made tremendous progress in the field of image denoising, most existing methods require huge training datasets and clean target images, which limits the application of these methods. In this paper, considering that clean target images are impossible to be obtained for X-ray pseudo-color images, we design a lightweight denoising method based on Noise2Void, which does not require clean target images. Moreover, a lightweight denoising network based on the depthwise separable convolution and U-net is designed to reduce computational consumption while guaranteeing denoising performance, thus ensuring wider applicability and adaptability. The method is evaluated on our real X-ray pseudo-color image dataset with the existing methods and the experimental results demonstrate that our method has better denoising performance and wider applicability.