基于改进横向中值滤波的南极磷虾声学数据去噪技术研究

Research on Denoising Technology for Antarctic Krill Acoustic Data Based on Improved Lateral Median Filtering

摘要: 南极磷虾作为全球渔业中资源量最大的生物，关系着极地渔业的捕捞效率和经济效益。南极磷虾的捕捞正逐步成为极地渔业发展中的热点问题。目前南极磷虾资源量的科学评估高度依赖于EK80获取的声学数据。然而在极地环境中，声波传播面临几何扩散衰减与海水吸收损失。若直接对未经校准的原始回波信号进行定量分析，深水区的微弱生物量信号极易被背景噪声掩盖，导致目标S_v被严重低估，进而引发资源量积分的量级偏差。针对上述衰减机理，本文提出基于现场水文特征的动态声学补偿与校准框架以及改进横向中值滤波与阈值保护去噪算法。定量分析表明，该动态校准模型有效消除了由于使用默认常数参数所引入的系统性偏差。相比于未经补偿的原始数据，本方法在200米典型磷虾栖息水深处减少了约4.8 dB的吸收衰减误差；在大于300米的深水区可排除20~30 dB的累积传输损失，使全局有效S_v值平均提升5~15 dB。

Abstract: Antarctic krill, as the largest biological fishery resource globally, is crucial to the harvesting efficiency and economic viability of polar fisheries. The harvesting of Antarctic krill is increasingly becoming a key issue in the development of polar fisheries. Currently, the scientific assessment of Antarctic krill biomass relies heavily on acoustic data acquired by Simrad EK80 systems. However, in polar environments, acoustic propagation suffers from significant geometric spreading loss and seawater absorption. Quantitative analysis performed directly on uncalibrated raw echo signals often leads to weak biological signals in deep water being masked by background noise, resulting in a severe underestimation of the target S_v and subsequent magnitude deviations in biomass integration. To address these attenuation mechanisms, this paper proposes a dynamic acoustic compensation and calibration framework based on on-site hydrological characteristics, integrated with an improved lateral median filtering and threshold protection denoising algorithm. Quantitative analysis demonstrates that the dynamic calibration model effectively eliminates systematic biases introduced by default constant parameters. Compared to uncompensated raw data, this method reduces absorption attenuation error by approximately 4.8 dB at a typical krill habitat depth of 200 m. In deep-water regions exceeding 300 m, it eliminates 20~30 dB of cumulative transmission loss, leading to an average increase of 5~15 dB in global effective S_v values.