Research On Underwater Multi-sensor Information Fusion And Recognition Algorithm Based On Submarine Pipeline

With the advancement of human science and technology and the increasing demand for resources,ocean exploration has become a new high ground for technological advances.Underwater sensing technology is the core task of ocean exploration.Among them,vision and acoustics are the most intuitive parts of underwater robot technology.Observation technology can effectively carry out underwater detection and target positioning and recognition,and is widely used in related fields such as underwater robot navigation,ocean exploration and underwater operations.Therefore,based on acoustic and optical information fusion for underwater targets The research of positioning perception technology has important scientific research value and is of great significance in line with national development strategies.This thesis studies sonar,optical camera and information fusion technology to realize the recognition and positioning of underwater targets and provide effective information for underwater detection and perception.The specific research contents are as follows:Firstly,in the information integration stage,the thesis studied the joint calibration method of sonar and camera.Firstly,the camera model is established.The camera parameters are calibrated by underwater imaging model,and the image is corrected by the obtained parameters.Then,the model of sonar frame and vehicle frame is established,and the model conversion relationship between them is obtained by joint calibration method.Secondly,the pre-processing of underwater images.Aiming at the problems of underwater and water quality causing imaging blurring,an improved DCP enhancement algorithm based on Dehaze Net is proposed.The transmission parameters are optimized through the Adaptive Bilateral Filter algorithm to obtain a more robust underwater enhancement algorithm,which can better Retain detailed information such as edges and textures,and perform performance verification for different underwater images.For the sonar image,it focuses on image segmentation,introduces common image segmentation,focuses on analyzing the MRF segmentation algorithm,and verifies its performance and reliability by collecting sonar image simulation.Thirdly,based on the multi-sensor information fusion method of particle filter fusion,the principle of particle filter is introduced.By establishing dynamic model and measurement model to fuse multi-mode data,overcome the problem of single sensor target positioning in a complex environment,and verify the algorithm through simulation feasibility.Finally,the underwater target recognition algorithm is studied.The deep learning framework is used to identify the sonar target,combined with CSPDense Net to improve the feature extraction layer in order to make full use of the inter-layer features,the DIo U-NMS non-extreme suppression algorithm,the Mish activation function can better improve the system's nonlinear performance,and the Drop Block regularization The method can better balance the generalization error and the empirical error,in order to obtain less calculation and more accurate prediction results,and perform simulation verification based on the acquired sonar image,and zoom the model based on the zoom cross-stage network.Get a network with better performance and accuracy.