Underwater Object Perception Techniques Based On Acoustic Image

Under the promotion of the marine rights and interests maintenance and marine power construction,the acoustic area security has become an important development direction of the marine acoustic technology,and the key is how to discern the underwater threat objects from the security areas quickly and accurately.Based on the underwater object perception framework,this paper focuses on underwater acoustic imaging,Region of Interest(ROI)detection and potential objects identification,adopting the method of theoretical analysis,simulation and experimental verification.In first layer,the underwater scene is mapped into acoustic images by means of multibeam water column imaging technology.The virtual beam interpolation and dynamic luminance allocation are applied to improve the quality of acoustic imaging,and an algorithm based on background estimation is proposed to suppress the sidelobe interference under high SNR.After the analysis of object representation for acoustic image,the local high order statistics(HOS)method is introduced to study the object representation in HOS space.In second layer,the strategy of subset censored is applied to the ROI detection,and the algorithm based on the integral image is used to improve computation efficiency.The fast subset censored constant false alarm rate detection(SC-CFAR)method is evaluated by the real acoustic image data.In addition,the ROI detection with SC-CFAR in local HOS space is studied.The original appearance of ROI can be restored as much as possible by the matched filter,and a optimized focusing effect of ROI can be obtained by the uncertain objective theory model,in the case of the unknown object size or number.In third layer,the study is carried out in two steps.The first step is to find the local feature suitable for the potential object in the acoustic image,and the second step is to use the local feature information to identify the the potential object.The common keypoints detection algorithms are applied to the acoustic image for the comparisons of detection performance,and the common feature descriptors are generated from the keypoints detected for the comparisons of matching performance.Then the keypoints detection algorithm and the feature descriptors are selected to form different local features.Depending on the strategy of detection before tracking(TBD),the method of local feature tracking is implemented in ROIs of the acoustic image sequences.The potential objects are thereby identified by the tracking statistics.Finally,experimental investigations on object perception are carried out,in which the diver is considered as a intruder.In order to simulate a variety of situations that may be encountered in acoustic area security,the intruder is designed to motion in multiple modes under three fields including fixed view of down-look,fixed view of front-look and mobile view of side-look,verifying the effectiveness of object identification under the underwater object perception framework.