Continuous gradient fusion class activation mapping: segmentation of laser-induced damage on large-aperture optics in dark-field images

Segmenting dark-field images of laser-induced damage on large-aperture optics in high-power laser facilities is challenged by complicated damage morphology, uneven illumination and stray light interference. Fully supervised semantic segmentation algorithms have achieved state-of-the-art performance but rely on a large number of pixel-level labels, which are time-consuming and labor-consuming to produce. LayerCAM, an advanced weakly supervised semantic segmentation algorithm, can generate pixel-accurate results using only image-level labels, but its scattered and partially underactivated class activation regions degrade segmentation performance. In this paper, we propose a weakly supervised semantic segmentation method, continuous gradient class activation mapping(CAM) and its nonlinear multiscale fusion(continuous gradient fusion CAM). The method redesigns backpropagating gradients and nonlinearly activates multiscale fused heatmaps to generate more fine-grained class activation maps with an appropriate activation degree for different damage site sizes. Experiments on our dataset show that the proposed method can achieve segmentation performance comparable to that of fully supervised algorithms.

Non-blind super-resolution reconstruction for laser-induced damage dark-field imaging of optical elements

The laser-induced damage detection images used in high-power laser facilities have a dark background,few textures with sparse and small-sized damage sites,and slight degradation caused by slight defocus and optical diffraction,which make the image superresolution(SR)reconstruction challenging.We propose a non-blind SR reconstruction method by using an exquisite mixing of high-,intermediate-,and low-frequency information at each stage of pixel reconstruction based on UNet.We simplify the channel attention mechanism and activation function to focus on the useful channels and keep the global information in the features.We pay more attention on the damage area in the loss function of our end-toend deep neural network.For constructing a high-low resolution image pairs data set,we precisely measure the point spread function(PSF)of a low-resolution imaging system by using a Bernoulli calibration pattern;the influence of different distance and lateral position on PSFs is also considered.A high-resolution camera is used to acquire the ground-truth images,which is used to create a low-resolution image pairs data set by convolving with the measured PSFs.Trained on the data set,our network has achieved better results,which proves the effectiveness of our method.

In-situ laser-induced surface damage inspection based on image super-resolution and adaptive segmentation method

In-situ laser-induced surface damage inspection plays a key role in protecting the large aperture optics in an inertial confinement fusion(ICF)high-power laser facility.In order to improve the initial damage detection capabilities,an in-situ inspection method based on image super-resolution and adaptive segmentation method is presented.Through transfer learning and integration of various attention mechanisms,the super-resolution reconstruction of darkfield images with less texture information is effectively realized,and,on the basis of image super-resolution,an adaptive image segmentation method is designed,which effectively adapts to the damage detection problems under conditions of uneven illumination and weak signal.An online experiment was carried out by using edge illumination and the telescope optical imaging system,and the validity of the method was proved by the experimental results.