几种提取内波波峰线的高精度检测方法分析

Analysis of Several High-Precision Detection Methods for Internal Wave Crest Line Extraction

海洋内波观测对于海洋学方面具有重要价值,海洋内波提取为进一步研究内波生成及预测提供了前提条件。本文就三种经典的传统图像处理方法及两种深度学习方法进行了比较,通过ERS-1、ENVISAT-1以及SENTINEL-1等合成孔径雷达遥感卫星获取的不同时刻、不同地点的数据来进行实验,采用F1分数、MACC及MIoU值验证不同方法的性能。通过实验发现传统图像处理方法与深度学习方法均能识别内波波峰线,其中Canny算法和自适应阈值方法更关注图像全局信息,光束曲线二叉树方法能够提取主要波峰线并且忽略一些细小且不明显的内波条纹,深度学习方法U-net与U2-net适应性强,在不同背景、噪声的情况都能提取较完整的内波波峰线。

Observation of internal waves in the ocean holds significant importance in the field of oceanogra-phy. The extraction of oceanic internal waves serves as a prerequisite for further research into their generation and prediction. This paper compares three classical traditional image processing meth-ods and two deep learning methods. Experiments were conducted using data obtained from Syn-thetic Aperture Radar (SAR) remote sensing satellites such as ERS-1, ENVISAT-1, and SENTINEL-1, captured at different times and locations. Performance evaluation was carried out using F1 score, accuracy (MACC), and Mean Intersection over Union (MIoU) values. Experimental results indicate that both traditional image processing methods and deep learning methods are capable of identi-fying the internal wave crest line. The Canny algorithm and adaptive threshold method prioritize global information in the images. The beam curve binary tree method can extract major internal wave crest lines while ignoring some smaller and less conspicuous internal wave patterns. The deep learning methods, U-Net and U2-Net, exhibit strong adaptability, successfully extracting compre-hensive internal wave crest lines under different backgrounds and noise conditions.