Research On Modeling The Correlation Between SAR Image Parameters And Target Detection And Recognition Performance

Synthetic Aperture Radar(SAR)target detection and recognition is a key link in achieving the application of SAR imaging technology,and changes in image quality have a direct impact on the performance of target detection and recognition,which in turn affects the application efficiency of SAR technology in fields such as battlefield reconnaissance and precision strike.Therefore,establishing the correlation between image quality and target detection and recognition performance,and quantitatively analyzing the constraint relationship between image quality and target detection and recognition performance,can provide theoretical basis for SAR imaging system design optimization,performance evaluation,and specific target recognition task resource allocation.Based on the above analysis,this thesis focuses on the selection of SAR image quality parameters and the correlation between image quality and SAR target detection and recognition performance.The main work is as follows:Analyzed the evaluation indicators of SAR image quality,identified the main factors affecting image quality,and designed image parameters from three aspects:ambiguity,noise level,and contrast.The effectiveness of the image parameters proposed in this thesis is demonstrated by calculating and simulating the image parameters of different quality SAR images,and comparing them with classical structural similarity parameters.The correlation between SAR image quality and target detection performance was studied.Two types of CFAR detectors based on Gaussian distribution andG~0distribution are used to detect SAR data with different image quality,and the intersection and union ratio between the detection results and the true value is used to describe the detection performance.Two representative data sets were selected,and the linear fitting method was used to establish the correlation model between the ambiguity,noise level and contrast and the SAR target detection performance.Through analysis,it was determined that the ground medium vehicle group(such as M-55 self-propelled howitzer)and the sea small ship group meet the requirements of high detection performance(intersection and merger ratio greater than 85%)and basic detection performance(intersection and merger ratio greater than 70%),The required blurriness,noise level,and contrast conditions for SAR image quality.The correlation between image parameters and target recognition performance was studied.Two representative datasets were selected and processed using single and double degradation methods.Target recognition experiments were conducted using Res Net18 based target recognition networks.Based on SAR image ambiguity,noise level,and contrast indicators,the recognition performance of the recognition network on degraded datasets was analyzed.A correlation model between image parameters and target recognition rate was established through linear regression,and the resolution and noise level required for high recognition performance(target recognition rate greater than 85%)and basic performance(target recognition rate greater than 60%)of small vehicle targets on the ground and large ship targets on the sea were determined through analysis.