基于哨兵二号遥感影像的近海养殖区提取方法研究

Research on extraction method of offshore aquaculture area based on Sentinel-2 remote sensing imagery

卫星遥感为近海养殖区信息的精确提取提供了有效的技术手段,对实现海水养殖业的科学监管具有重要意义。现有光学遥感手段对养殖区信息的提取多侧重于影像空间特征的使用,对于光谱和空间信息的结合在养殖区特征表达与判别提取方面的综合利用还有待进一步挖掘。本文利用哨兵二号卫星遥感影像,针对近海养殖区小样本、多维度、水体组分复杂等特点,提出了一种光谱和空间信息结合的养殖区提取算法。首先通过指数计算增强养殖区目标的特征表达能力;在此基础上采用支持向量机模型进行光谱信息的初分类;最后结合马尔科夫随机场模型对初分类结果进行后处理,综合利用谱空信息实现近海养殖区信息的精准提取。该算法在研究区达到了94.46%的整体分类精度,相较于仅利用光谱信息和纹理特征增强的方法,各类养殖区的提取精度均有明显提升,提取结果中海水与浮筏的混分现象也得到了有效改善。本文算法模型为近海养殖区的自动提取提供了一种新思路。

Satellite remote sensing provides an effective technical means for accurately extracting information from offshore aquaculture areas, and is of great significance to the realization of scientific supervision of the marine aquaculture industry. Existing optical remote sensing methods mostly focus on the use of image spatial features in the extraction of aquaculture information, and the comprehensive utilization of the combination of spectrum and spatial information in the feature expression and discrimination extraction of the aquaculture area needs to be further explored. In this paper, using the remote sensing image of Sentinel-2 satellite, aiming at the characteristics of small sample, multi-dimension, and complex water composition in offshore aquaculture area,this paper proposes an aquaculture area extraction algorithm that combines spectral and spatial information.Firstly, index calculation is performed to enhance the feature expression ability of the target in the aquaculture area;on this basis, the support vector machine model is used to perform the initial classification of the spectral information;finally, combine the Markov random field model to post-process the initial classification results,and comprehensive use of spectral and spatial information to achieve accurate extraction of offshore aquaculture area information. This method has achieved an overall classification accuracy of 94.46% in the study area, Compared with the method that only uses spectral information and texture feature enhancement, the extraction accuracy of various aquaculture areas has been significantly improved. In the extraction results, the mixing phenomenon of seawater and floating rafts has also been effectively improved. This algorithm model provides a new idea for the automatic extraction of offshore aquaculture areas.