Aero-optical Transmission Effect Correction Method Based On Lightweight Network Fuzzy Kernel Estimation

When the aircraft is flying at high speed,the image of the detector is blurred and degraded due to the influence of atmospheric turbulence.This phenomenon is called aero-optical transmission effect.The degradation of the detector's imaging has a great impact on the aircraft's target detection and navigation and guidance functions,so it is of great significance to study the corresponding restoration and correction methods.Aiming at the problem of aerodynamic transmission effect restoration and correction under high-speed aircraft embedded system,due to its low computing power and real-time requirements,this thesis proposes a method of aerodynamic transmission effect correction combining a lightweight parameter estimation network and FTVd non-blind defuzzification algorithm.The main research contents of the thesis are as follows:First,the commonly used aerodynamic transmission effect correction algorithm is evaluated from two aspects of restoration effect and algorithm real-time performance.According to the evaluation result,the FTVd non-blind defuzzification algorithm is finally selected,and the robustness of the algorithm is analyzed.Then,a parameter estimation algorithm based on lightweight network is proposed.The use of a deep separable convolution structure for parameter estimation network design reduces the amount of parameters by 87.3% and the amount of calculations by82.8% while achieving the same prediction accuracy.Before prediction,the fuzzy image is cropped,several fixed-size fuzzy sub-images are obtained,and small-scale sub-images are input into the network for prediction,which further reduces the amount of parameters and calculations.After the prediction,the predicted values of several sub-pictures are averaged,the predicted average value is used as the prediction result of the blurred image,and the prediction result is compensated to reduce the prediction error of the network.The final network prediction error of the fuzzy kernel standard deviation in the interval[1,9] is between [-0.25,0.25].Finally,the restoration effect of the aerodynamic transmission effect correction method proposed in this thesis is verified by simulation experiments.In the Gaussian blur simulation images with standard deviations of [1,9],the PSNR values of the restoration results of lightly blurred,moderately blurred and severely blurred images have increased by 2.23 d B,1.52 d B and 1.27 d B,and the SSIM value has increased by 0.19,0.12 and 0.05.For real blurred images,the method in this thesis also has a good restoration effect.