ANLResNet网络的前视声呐图像散斑去噪方法  

Despeckling for Forward Looking Sonar Image Based on ANLResNet

作　　者：陈朋 徐泽楠 赵冬冬 郭新新 CHEN Peng;XU Ze-nan;ZHAO Dong-dong;GUO Xin-xin(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China;College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China;Institute of Deep-sea Science and Engineering Chinese Academy of Sciences,Sanya 572000,China)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310023 [2]浙江工业大学信息工程学院,杭州310023 [3]中国科学院深海科学与工程研究所,海南三亚572000

出　　处：《小型微型计算机系统》2022年第2期355-361,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金青年科学基金项目(62001418)资助;浙江省自然科学基金项目(LQ21F010011)资助;浙江省属高校基本科研业务费专项资金项目(RF-C2019001)资助;中国科学院战略性先导科技专项项目(A类)(XDA22030302)资助。

摘　　要：前视声呐(Forward-Looking Sonar,FLS)使用换能器基阵收发声波,通过回波探测水下物体.在浅海环境,由于水下介质的反射、散射与不均匀波动,前视声呐图像极易引入散斑噪声.本文针对前视声呐图像散斑噪声,结合SRResNet与非对称金字塔非局部块,提出了ANLResNet网络用于前视声呐图像去噪,并针对前视声呐图像特性,使用FieldⅡ构建模拟前视声呐图像数据集,对网络进行训练.实验结果表明,本文提出的ANLResNet网络能有效的去除前视声呐图像中的散斑噪声,获得良好的视觉效果.并通过峰值信噪比(Peak Signal to Noise Ratio,PSNR)、等效视数(Equivalent Number of Looks,ENL)、散斑抑制指数(Speckle Suppression Index,SSI)3个图像质量评价指标评价降噪效果.在模拟前视声呐图像去噪上,本文算法相比于传统算法、改进BM3D算法和Autoencoder网络,平均PSNR至少提高了8.12%.在真实前视声呐图像去噪上,本文算法相比于传统算法、改进BM3D算法和SRResNet等效视数至少提高了16.77%,散斑抑制指数至少降低了2.84%.相比于Autoencoder网络等效视数提高了4.30%.本文方法主要用于前视声呐图像去噪,对于其他声学图像的降噪,散斑噪声的抑制上也有一定的应用价值.Forward-Looking Sonar(FLS)uses transducers to send and receive sound waves and detect underwater objects through echoes.In the shallow sea environment,due to the reflection,scattering and uneven fluctuation of the underwater medium,the forward-looking sonar image can easily introduce speckle noise.Combining SRResNet and Asymmetric Pyramid Non-local Block,the ANLResNet network is proposed to filter the multiplicative speckle noise in forward looking sonar images.According to the characteristics of forward looking sonar images,a simulated forward looking sonar image data set is constructed by the FieldⅡto train the network.The experimental results show that the ANLResNet network proposed in this paper can effectively remove the speckle noise in the forward looking sonar image and obtain good visual effects.Moreover three image quality evaluation metrics including the Peak Signal to Noise Ratio(PSNR),equivalent noise(ENL),speckle reduction index(SSI)are used to evaluate the noise reduction effect.Then compared with the traditional algorithm,the improved BM3 D algorithms and the Autoencoder,the average PSNR of the proposed algorithm is at least 8.12%higher in the denoising of the simulated forward-looking sonar image.Compared with the traditional algorithm,the improved BM3 D algorithms and the SRResNet,the ENL of the proposed algorithm is at least 16.77%higher in the denoising of the real forward-looking sonar image,and the SSI is reduced by at least by 2.84%.Compared with the Autoencoder,the ENL of the proposed algorithm is 4.30%higher in the denoising of the real forward-looking sonar image.The proposed method is mainly used for the denoising of forward looking sonar images.It also has certain practical value in the denoising of other acoustic images and the suppression of speckle noise.

关 键 词：前视声呐 散斑噪声 图像降噪 FieldⅡ SRResNet 

分 类 号：U666.7[交通运输工程—船舶及航道工程] TP391.41[交通运输工程—船舶与海洋工程]