Research And Software System Implementation Of Integrated Diagnosis And Detection Method Of SAR Equipment Based On MTL-CNN

As the core equipment in the field of modern information reconnaissance,Synthetic Aperture Radar(SAR)has been paid more and more attention by researchers from all over the world.Due to the complex operating environment and the failure of its own components,SAR equipment is prone to failure of reconnaissance missions,so it is of great significance to the research on diagnostics of SAR equipment.The existing SAR diagnostics is composed of two parts serially: adaptive evaluation and failure cause reasoning,which has the problems of low efficiency and low accuracy of diagnostics.In this thesis,related research is carried out on this problem.Firstly,the adaptive joint noise reduction algorithm based on non-subsampled shearlet transform(NSST)is used to preprocess the SAR image,and then the multi-task guidance network MTL-CNN is used to realize the comprehensive diagnosis and detection of SAR equipment.Combined with the application requirements,an efficient SAR equipment integrated diagnostic and detection software system is finally constructed.The main research contents of this thesis are as follows:(1)Aiming at the problem that the diagnostics results of SAR equipment are greatly affected by speckle noise in the actual scene,this thesis deals with high and low frequency noises on the basis of NSST decomposition,and then constructs an adaptive joint noise reduction algorithm AH-NSST.For the high frequency part,considering the large size of SAR images,this thesis proposes an adaptive cyclic window Bayesian threshold shrinkage algorithm.For the low frequency part,this thesis adopts the bilateral filtering algorithm of variable scale window for noise reduction.Compared with the existing algorithm on the benchmark dataset and the real SAR dataset,the results of the experiment indicate that the noise suppression index of the algorithm in this thesis is improved by 9.6% on average,and the edge preservation index is improved by 8.7% on average.(2)Aiming at the problems of low efficiency and low accuracy of the existing diagnostics algorithms and being affected by the terrain environment,this thesis combines the idea of multi-task learning,introduces auxiliary tasks and designs a guided shared learning module,thereby constructing a task-guided learning SAR equipment diagnostic model MTL-CNN.At the same time,this thesis use a multi-scale module in the modelbuilding process to enhance the feature extraction capability of the model,and design a dynamic weight loss function to optimize the training process,thereby further improving the performance of the model.Through the module ablation experiment of the model and the comparison experiment with the existing algorithm,it can be obtained that the comprehensive indicators of the algorithm in this thesis in adaptability evaluation and failure cause reasoning are improved by 4.4% and 0.4% respectively,and the terrain generalization ability is improved by 15.5% and 17.5% respectively.(3)Aiming at the actual application scenarios of airborne SAR equipment diagnostics,this thesis analyzes the requirements of the diagnostics system,and combines the above denoising and diagnostics models to build an integrated SAR equipment diagnostics software system which can provide stable and efficient diagnostic service for users.The system also implements basic functions such as user management,experiment management,and historical data management.By showing the functional test and performance test results of the software system,it can be obtained that the execution time of a single diagnostics algorithm of the system is less than 1min,which verifies that the software system meets the needs of diagnostics.