黄渤海典型鱼类的宽频声散射特性与种类识别

Acoustic survey is an important method for fishery stock assessment, in which species identification is a key issue for the partition of total echo-integration. Traditionally, the species composition is usually determined by trawl sampling. However, the trawl sampling is not very accurate due to low spatio-temporal sampling rate coupled with fish avoidance and escapement. Therefore, how to improve the accuracy of fish species identification and classification is an urgent issue for accurate acoustic-based fishery stock assessment. The acoustic backscattering from a fish over a broad frequency range can provide various characteristic information, and the broadband technique is entering practical application, which is a promising way for the identification and classification of multiple species. Aiming for the species identification and classification during stock assessment which is the urgent demand of quota management in China, the project will construct the 3D finite-element analytic models of acoustic scattering from several representative fishes in the Yellow and Bohai Sea by ray scanning and measurement of physical parameters. Then the broadband acoustic scattering properties of the fishes and the effect of length, tilt angle and acoustic impedance will be numerically studied by using finite-element and boundary-element coupled method. Meanwhile, the differences of spectral characteristics scattering from different fish species will be figured out. Finally, a broadband based method for specific species identification will be exploited and validated by in-situ experimental data. The research will prompt the academic subject development of fisheries acoustics in China, and accelerate the application of broadband techniques in the acoustic survey of fishery resources.

基于回声积分技术的声学调查已成为渔业资源定量评估的主要手段，其中鱼种识别是各种类积分值分配的关键。传统鱼种识别主要依靠拖网取样，而受时空采样率限制和网具选择性等因素影响，拖网得到的种类组成并不十分准确。因此，如何提高种类识别的准确性是声学评估亟待解决的主要难题之一。鱼类宽频回波包含丰富的目标特征信息，且宽频技术正进入实际应用阶段，为实现鱼种识别提供了一种全新途径。本研究针对我国多种类混栖情况下渔业资源准确评估及限额管理的迫切需求，选取黄渤海典型鱼类为研究对象，结合射线扫描和声学参数测量，构建各鱼种的标准三维有限元散射模型；基于有限元-边界元方法对其宽频声散射特性进行数值仿真研究，探讨鱼类体长、倾角及声阻抗等参数变化对频谱特征的影响，摸清不同鱼种的宽频散射特性差异；建立特定鱼种的宽频声学识别技术，并进行海上现场验证。该研究将促进我国渔业声学学科发展，推动宽频技术在渔业资源调查的实质性应用。

基于回声积分技术的声学调查已成为渔业资源定量评估的主要手段，其中鱼种识别是各种类积分值分配的关键。传统鱼种识别主要依靠拖网取样，而受时空采样率限制和网具选择性等因素影响，拖网得到的种类组成并不十分准确。鱼类宽频回波包含丰富的目标特征信息，且宽频技术正进入实际应用阶段，为实现鱼种识别提供了一种全新途径。本研究针对多种类混栖情况下渔业资源评估及限额管理的迫切需求，以鳀鱼、带鱼、银鲳、灯笼鱼、秋刀鱼、沙丁鱼和南极磷虾这几种重要经济和生态物种为研究对象，构建了各鱼种的三维声散射模型；基于有限元-边界元方法（FEM-BEM）对其宽频声散射特性进行仿真研究，分析了体长、倾角等参数变化对频率响应特征的影响，了解不同鱼种的宽频散射特性差异，探讨了基于宽频散射特征进行种类识别的可行性；基于回声映像特征，探究了利用机器学习算法进行特定鱼种识别的新方法，并针对具有显著集群行为的鳀鱼开展了应用研究。研究发现，FEM-BEM解决了MB-DCM中低频、两端入射时的准确性问题；对于弱散射目标，FEM-BEM与DWBA模型高度吻合。鱼类不同游泳倾角下的目标强度可相差几十分贝，有效平均目标强度随频率升高而降低；无鳔鱼类的目标强度远低于有鳔鱼类，且两者的宽频散射特征也有明显差别，为基于散射频率响应的声学识别提供了依据。而具有不同几何形态的不同有鳔鱼类的宽频散射特征差别不明显，其声学识别方法尚需进一步研究。基于回声映像特征和机器学习算法的鳀鱼识别方法估算的资源量比实际高20.2%，应用效果良好，未来可通过增加训练样本数、考虑鱼类散射频率响应等特征量，提高识别的准确率。本研究引入的FEM-BEM方法基本可以满足无鳔鱼类、有鳔鱼类和浮游动物等海洋生物的声散射求解，有望成为海洋生物声散射研究的标准方法；提出的鱼类声学识别方法无需人工取样，具有高效、生态的特点，具有良好的实际应用价值，将推动新技术、新方法在渔业资源声学探测评估的应用。