Multi-target Autonomous Detection Method Based On Single Vector Hydrophone And Realization Of GPU Software

Marine safety problems have been an important part of ocean strategy.Ocean target detection is developing forehead constantly.Because acoustic vector sensor has directivity characteristics which have no relation with frequency,it has get wide attention in the area of target bearing estimation.There is an absolute advantage for GPU on single-precision floating-point computing power,external memory bandwidth,and parallel computing power,which will surely lead the future of high-performance computing.Firstly,the dissertation deduces the correlation between the sound pressure and particle velocity in the isotropic noise field and coherent source radiation sound field.And this establishes the basis of estimating the azimuth for the average sound power amplifier in isotropic interference fields.In order to study the influence factors of azimuth estimation of line spectrum,the time-domain average sound power and the average frequency domain sound power are respectively simulated and analyzed,that test and verify the line spectrum azimuth estimation can well achieve the purpose of multi-target resolution.According to the feature information radiated by noise of line spectrum target such as ship,a multi-target line spectrum multi-target autonomous detection and tracking method is designed.And the multi-target continuous detection and tracking method based on multi-level detection is designed.The anti-interference performance and multi-target resolution of the continuous spectrum azimuth estimation are simulated.Secondly,According to the characteristics of the target detection algorithm,analysis the parallel algorithm including the intensive data calculation module.In-depth study of the basic principles of FFT transform,after analyzing the highly parallelizable features in detail,combined with the characteristics of CUDA multi-threaded operating mode,we study the implementation of FFT parallel algorithm design and complete algorithm optimization.The basic principle of FIR filter and the characteristics of overlap reservation method are analyzed and studied.Combined with the characteristics of CUDA asynchronous transmission,a parallel algorithm of FIR filter is designed.And then after analying and comparing the target detection serial algorithm and parallel algorithm,using lake test data to complete the serial program and parallel program task execution.Finally,The performance of the algorithm and software were verified and evaluated by field experiments.The results verify that automatic detection of line spectrum signals and automatic detection of continuous spectrum signals are effective and reliable.