A denoising method based on CNN for the measurement of far-field focal spot of high power laser using Schlieren method

Abstract

Schlieren method is an effective method to measure the high dynamic range far-field focal spot of high power laser at present, it has been used in the comprehensive diagnosis system of the host facility of the national large scientific device. However, the background subtraction algorithm is used for denosing process of the main lobe and side lobe images in the experiment, which could not effectively remove the noise under the condition of strong radiation. In order to improve the reconstruction accuracy and reliability of far-field focal spot measurement based on schlieren method, there are two aspects have been proposed and optimized in this paper as follows, the mathematical model for the measurement of high dynamic range far-field focal spot is constructed, the noise of images captured by detected CCD is removed effectively. Firstly, the mathematical model of laser focal spot measurement is constructed, which provided a theoretical basis for the measurement of far-field focal spot based on schlieren method. Secondly, the denoising algorithm based on Convolutional Neural Network (DnCNN) is introduced into the denoising of main lobe and side lobe CCD image, which can effectively remove the noise of different levels (0-75db) of main lobe and side lobe CCD image.The experimental results show that the noise signal of main lobe and side lobe images are removed by using DnCNN algorithm, and the best reconstructed image of far-field focal spot of high power laser is obtained when the influencers of amplified noise of main lobe image is reduced greatly. After the above optimization and improvement, it will meet the requirements for accuracy and efficiency of the measurement of high dynamic range far-field focal spot.