Filtering And Segmentation Of Sonar Images Based On Pulse Coupled Neural Network

Regardless of manual interpretation or automatic recognition,the speckle noise suppression of sonar images is necessary because speckle noise is severely affected by artificial interpretation and automatic recognition of images.The sonar image segmentation is the most critical step before the sonar image is filtered and the segmentation can extract the features needed for the recognition.There are obvious target bright area,target dark area,and reverb zone in the sonar image,dark spots(dark speckle noise)in the target bright area,as well as bright spots(bright speckle noise)in the target dark area.Combining with the characteristics of the image and the mechanism of the Pulse Couple Neural Network(PCNN)that makes the neurons of the pixels with similar adjacent pixel values in the image impulse simultaneously,the threshold function in the PCNN neuron model is able to be improved.This dissertation uses the improved PCNN to filter and segment sonar images.In the section of filter,firstly,setting the appropriate parameters to make the improved PCNN iterative convergence,and then the above three areas are respectively found and the dark spots in the target bright area and bright spots in the target dark area are found.Finally,using the modified median filter to filter out the dark spots in the target bright area and the bright spots in the target dark area respectively.In the section of segmentation,the improved PCNN neuron model is also used in this dissertation.The three regions are set to three different gray values after respectively finding the above three regions,utilizing the PCNN iterative convergence,after that,segmentation is completed.In the end,four groups of sonar images are filtered and segmented respectively using the improved PCNN neuron model.In the filtering experiment,the improved PCNN filtering algorithm of this dissertation is compared with median filtering,Frost filtering,Kuan filtering,Lee filtering and two filtering algorithms that use PCNN technology.In the segmentation experiment,the the improved PCNN segmentation algorithm of this dissertation is compared with the Kapur two-dimensional double-threshold method,the Otsu double-threshold method,and a segmentation algorithm that uses PCNN technology.Finally,the effects of filtering and segmentation are analyzed from the perspective of the target light area and the target dark area,and it is concluded that the results of the improved PCNN filtering and segmentation algorithm developed in this dissertation are better than other algorithms and they are suitable for sonar images with obvious bright and dark areas.The difference between the filtering and segmentation adopting the PCNN neuron model is that the parameters of the threshold function is different.