Research On Integrated Optomechanical Analysis Of Aerial Camera Based On Neural Network

As a kind of analysis method which considers optics,structural mechanics and thermodynamics,integrated optomechanical analysis has been widely used in the design stage of optical instruments.There is a large temperature difference between the working environment and the setting environment of aerial camera.It is necessary to use the integrated optomechanical analysis to predict the temperature adaptability,so as to ensure that the image quality is not affected by the large temperature difference.In the process of analysis,firstly,finite element analysis software is used to establish the finite model of aerial camera,and the displacement data of optical element surface nodes are obtained by applying temperature load.Then,Zernike polynomial is used to fit the data,in order to solve the polynomial coefficient of the surface deformation caused by heat.Finally,the solved coefficient is imported into the optical design software.The effect of temperature load on the image quality of optical system can be quantitatively analyzed.As an interaction interface between optical and mechanics data,the solving accuracy of Zernike polynomial coefficient directly determines the results of integrated optomechanical analysis.The existing Zernike coefficient solving algorithms mainly include: least square method,Gram-Schmidt orthogonal method,Householder transformation method and singular value decomposition method.For large-scale data operation,the above traditional algorithms may have ill-conditioned problems,which may lead to the decrease of fitting accuracy,or the high computational complexity severely limits the solving efficiency.To solve the above problems,this paper proposes a Zernike polynomial coefficient solving algorithm based on one-dimensional convolutional neural network,and applies it to the integrated analysis of optomechanical systems of the short-wave infrared aerial camera.It focuses on the completion of the optomechanical system design,thermal elastic analysis of optomechanical systems,node displacement data fitting,optical performance evaluation and other research work,which can be summarized as follows:(1)The integrated optomechanical analysis process is elaborated,and the finite element analysis theory and thermoelasticity theory are studied.The transformation algorithm of node displacement data into vector height data and vector height fitting algorithm based on Zernike polynomial are presented.Four algorithms for solving Zernike polynomial coefficients are analyzed,and the advantages and disadvantages of these algorithms and their application scenarios are discussed.(2)The general design index of the short-wave infrared aerial camera is drawn up,and the characteristics of the transmission optical structure,the reflection optical structure and the refraction optical structure are compared.With the Ritchey Chretien(RC)system as the initial structure,the design of the refraction optical system and the optical-mechanical structure are completed.The photomechanical finite element model of aerial camera is constructed by using finite element analysis software,and the thermoelastic analysis is carried out by applying corresponding thermal conditions.The thermal node displacement data of the main mirror is extracted to construct the data set required for training the neural network model.(3)The Zernike coefficient algorithm based on one-dimensional convolutional neural network is studied.A one-dimensional convolutional neural network model is constructed,and the model is optimized by using residual connection structure.In the process of feature transfer,batch normalization is introduced and global average pooling is used to replace the full connection layer.In order to train a network model equivalent to the variation of vector height,optimization methods such as batch normalization,Dropout and regularization are used to alleviate the overfitting problem.Adam adaptive learning rate optimization algorithm is used to achieve rapid convergence of the network model.(4)The one dimensional convolutional neural network is trained using the thermal node displacement data of the primary mirror node,and the Zernike coefficient of the optical surface is solved.According to the results of coefficient solving,the optically deformed surface is fitted,and the residual difference between the fitting vector height data and the finite element analysis vector height data is calculated,which proves the effectiveness of one-dimensional convolutional neural network.The Zernike coefficient of the optical surface deformation of the primary mirror is introduced into the optical design software and the influence of temperature change on the optical performance of short-wave infrared aerial cameras is quantitatively evaluated by the ray tracing method.