Research On Recognition Method Of Photon Imaging For Marine Target

With the increasingly fierce frictions at sea,the importance of maritime security has drawn wide attention.The photon radar detector has the sensitivity of single photon and can receive the imaging of weak echo signal at a long distance.Its effective working distance is far beyond that of conventional lidar,but the photon model point cloud obtained is very sparse and contains a lot of dense noise.The key research problem of photon imaging target recognition lies in photon denoising and feature extraction.In this paper,airborne photon radar is used to detect and identify long-distance ships at sea as the research background,and the research is carried out from three parts: photon noise data processing,maritime detection background segmentation,target feature extraction and recognition.(1)Due to the loss of some signal points in point cloud denoising based on density,this paper proposes a new point cloud denoising method based on the distribution characteristics of push-sweep point cloud data.The airborne photon detection data are divided into different intervals along the direction of the laser beam,and the maximum density point of each interval is calculated as the center point.The average length of some intervals is set as the threshold value,and the points whose distance from the center point is greater than the threshold value are filtered out.Finally,the improved statistical filtering method is used to smooth the data and complete the final denoising.MABEL data test results show that the proposed algorithm can retain more than 98% of signal point cloud while denoising.(2)In this paper,the characteristics of common field scenes on the sea are studied.Combined with the scanning accuracy of photon radar,3d point cloud models of sea surface,different types of ships and reefs are obtained by modeling software,and the scene model of photon detection on the sea is established.According to the research background of this paper,the traditional RANSAC plane-fitting algorithm and DBSCAN algorithm are improved to realize the filtering of the sea surface point cloud and the clustering of the remaining field scenes.Finally,a simple binary classification algorithm is proposed according to the shape features.By fitting the plane of the ship model and the reef model,the corresponding features are extracted to realize the extraction of ship targets.(3)After obtaining the 3D model of the ship through software simulation,11 ship samples in different directions are obtained through cutting;The dimensional features,geometric features,bow Angle and normal vector histogram of each ship sample are proposed and calculated.The optimal dimension of features is obtained through simulation experiments.Finally,target recognition is carried out with random forest classifier,and the simulation results show that the accuracy of target recognition is more than 98%.Based on the equivalent experiment of airborne lidar,the 3d point cloud model of measured ship on river is obtained,and the features are extracted for recognition experiment,which verified the effectiveness of the proposed recognition algorithm.