基于视频图像获取冰面特征的自动检测算法研究

Research on automatic detection algorithm based on video image acquisition for ice surface feature

基于船基或岸基平台拍摄视频图像,来获取冰凌块体几何参数的方法越来越成熟,而冰凌块表面特征信息提取还在发展之中。其技术难点在于如何自动化地处理批量图像,以提高工作效率。本文以船基图像中提取北极海冰表面融池分布信息参数为例,首先讨论了阈值法、K-means聚类算法、分水岭分割、支持向量机和随机森林算法提取冰面特征值的优缺点;然后对海冰表面特征检测开源算法OSSP进行改进,成功地将其应用到区分海冰图像中的冰雪、开阔水域与融池等表面特征;最后利用北极现场海冰图像及其他方法处理的结果进行对比,验证改进方法自动化地处理图像的可行性。发现本改进方法是自动提取船基海上流冰和岸基河冰流凌视频图像中冰面特征参数的一种潜在新方法。

The methods of capturing geometric parameters of ice blocks based on video images taken from ship-based or shore-based platforms become more and more mature,while the way to extract surface feature information of ice blocks is still under development.The technical difficulty lies in how to automatically process batch images to improve work efficiency.The advantages and disadvantages of several methods for extracting eigenvalues of sea ice surface,such as thresholding method,K-means clustering algorithm,watershed segmentation,support vector machine and random forest algorithm,were discussed.Then,the open source sea ice surface feature detection algorithm OSSP was improved and successfully applied to distinguish surface features such as snow and ice,open water and melt pond in sea ice images.Finally,the results of Arctic sea ice images and other methods were compared to verify the feasibility of the improved method in automatic image processing.It was found that the improved method was a potential new method for automatic extraction of ice surface characteristic parameters from video images of ship-based sea ice flow and shore-based river ice flow.