Research On Underwater Target Recognition Method Based On Environmental Noise Model

To ensure the safety of shallow water areas,which are concentrated with ports,docks,anchorages,and shipping routes,equipping security systems with underwater protection capabilities has become an important means of water safety.Underwater target recognition systems are an important component of underwater security systems.However,during underwater target recognition,the accuracy of target identification is greatly affected by interference from irrelevant signals such as underwater environmental noise.Therefore,this thesis focuses on researching how to use underwater environmental noise to improve underwater target recognition methods.Firstly,considering the diversity and complexity of underwater environmental noise,noise models are established for each noise source and then superimposed.The noise models are simplified and analyzed according to the composition of noise sources in the experimental environment.Noise characteristics are obtained by modeling noise in different depth water layers,and the established noise models are evaluated using collected noise data.The R squared indexes are all higher than 0.9 under the condition of no external interference.The results show that the noise models have good reliability and stability and can be used as the consistency estimation of the actual noise characteristics.Furthermore,in the process of target echo signal recognition,the received echo signals are separated from noise by using variational mode decomposition and independent component analysis,and the decomposition layer number is optimized using the noise characteristics.The multi-scale features of the target are extracted by wavelet packet transform and sparse decomposition,and the feature vectors are constructed.Then,the support vector machine algorithm based on particle swarm optimization is used to achieve target signal recognition.The experimental results show that the recognition accuracy of different target signals processed by the algorithm proposed in this thesis is above 90%,and reduce the computational complexity,which verify the effectiveness of the signal recognition method proposed in this thesis.Finally,in target sonar image recognition,the image resolution of the target is improved by the modified guidance vector optimization beamforming algorithm.Then,the guided filtering algorithm is used to denoise the sonar image,and the edge-aware factor is constructed based on the noise characteristics to optimize the filtering algorithm for better edge preservation performance.The denoised sonar image is then processed by the Yolov5 s algorithm improved based on attention mechanism for target recognition.Simulation results show that the optimized algorithm improved the three target recognition accuracy by 5.5%,5.02% and 3.56%,and reduce the frame image average processing time by 1.78 ms,thus verifying the effectiveness of the algorithm.