Underwater Target Detection And Recognition Based On Sonar Image Feature Extraction

Underwater sonar detection is the main way of ocean long-distance detection at home and abroad,and the processing of underwater imaging has always been a research hotspot with application value and challenge.After acoustic imaging,the image itself is not as high as the resolution of the traditional optical image,and the acoustic imaging quality is worse due to the complex underwater environment.There are artifacts and complex background noise,and the processing effect of the traditional target detection and recognition algorithm is not satisfactory.Therefore,in this paper,the characteristics of underwater acoustic image are studied,and a variety of improved algorithms for sonar image processing are proposed.The noise reduction preprocessing of underwater acoustic image is the premise of sonar image target detection and recognition.Due to the complexity of noise in acoustic image,the classical methods of optical image noise reduction have limited effect on its processing.In this project,we use the unsampled contour wave transform to decompose the acoustic image to get the high and low frequency sub-bands.On the basis of this,we simplify and improve the parameters of the pulse coupled neural network model and fuse the high and low frequency images.The reconstructed acoustic image retains the edge information of the acoustic image,while removing the speckle noise.Even when the noise level is gradually increased,the average signal-to-noise ratio of the method proposed in this paper is 34.8%,14.5%,and 9.4% higher than that of frequency domain noise reduction,double tree and double density noise reduction,Lee filtering,respectively,which has higher anti noise performance.A large number of data from sonar detection in wide water area brings some difficulties to target recognition.Before target recognition and matching,it is necessary to judge whether the target of acoustic image has any redundancy calculation in the later stage.In this paper,twodimensional empirical mode decomposition is used to decompose and reconstruct sonar image,and detect the envelope curve of the extracted difference extreme point.Experimental analysis shows that using the difference instead of the ratio in the detection of long and short time windows can better highlight the fluctuation amplitude change of the envelope curve of the acoustic image containing the target.The method is simple and the detection accuracy is up to 97.2%.It is a key problem to identify the target in underwater acoustic image processing.Most of the target matching methods are trying to reduce the mismatch rate of feature points.In this paper,we use the idea of finding the best feature point pairs in the target area of acoustic image: using the improved CFAR technology,we can select the feature point pairs in the target area from a large number of feature point pairs matched by surf algorithm.At the same time,combined with the statistical feature detection of the ring-shaped region,judge whether the two match,calculate the geometric relationship of the feature point pairs after the matching is successful,and find the matching target area after the rotation correction.The experimental results show that this method can effectively resist the influence of target deformation and rotation under the complex background of acoustic image,and the target matching overlapping rate is increased by 16.57% and 32.59% respectively compared with the best brother similarity algorithm and its improved algorithm.Finally,the algorithm of target detection and matching is implemented in Python environment,which makes full use of its powerful resource environment.Open CV image visual library and pyqt5 control set are called to complete the design of visual graphic interaction system,which is compatible with all operating platforms and has high portability.