一种藻类养殖区自动化提取的Otsu优化算法

Automatic Extraction of Algae Aquaculture Area Based on an Optimized Otsu Algorithm

针对传统Otsu算法在藻类养殖区分布信息的自动化提取过程中存在欠/过分割、计算量大和运算效率低等问题,提出一种优化的藻类养殖区自动化提取Otsu算法(GA-Otsu)。GA-Otsu算法在最大类间方差的基础上,引入类内方差,共同参与阈值选取,提高藻类阈值选取的准确性,并用遗传算法代替遍历法快速搜索最优解,实现藻类养殖区分布信息的准确、快速、自动化提取。选取三沙湾为研究区,综合利用同时期不同空间分辨率的两种遥感影像(Sentinel-2 MSI与GF-2),基于影像的光谱特征和敏感波段分析,用比值运算构建一个藻类光谱指数(algal spectral index,ASI),并运用GA-Otsu算法实现藻类养殖区自动化提取。GA-Otsu算法运用在Sentinel-2影像上,总体精度提高4.74%,Kappa系数提高0.14;而运用在GF-2影像上,效果提升得更加明显,总体精度提高10%左右,Kappa系数提高0.18。实验结果表明GA-Otsu算法不受传感器性能的影响,对不同数据源仍具有一定的普适性。此外,GA-Otsu算法运算量大幅减少,时间效率提高了85%,具有较高的实际应用价值。

An optimized algorithm(GA-Otsu)for automated extraction of algae aquaculture area is proposed for the problems of under-and over-segmentation,large computation and low operational efficiency of the traditional Otsu algorithm in the process of automated extraction of algae aquaculture area distribution information.The algorithm introduces intra-class variance on the basis of maximum between-class variance.It jointly participates in threshold selection to improve the accuracy of algae threshold selection.And the genetic algorithm is used to replace the traversal method to quickly search for the optimal solution to achieve accurate,fast and automatic extraction of algae aquaculture area distribution information.In this paper,Sansha Bay is selected as the study area,and two remote sensing images(Sentinel-2 MSI and GF-2)with different spatial resolutions in the same period are used comprehensively.Based on the spectral features and sensitive band analysis of the images.An algal spectral index(ASI)was constructed using a ratio operation,and an optimized Otsu algorithm(GA-Otsu)was applied to automate the extraction of algal aquaculture areas.The algorithm is applied to Sentinel-2 MSI images,with an overall accuracy improvement of 4.74%and a Kappa coefficient improvement of 0.14.On GF-2 images,the effect is more obvious,with an overall accuracy improvement of about 10%and a Kappa coefficient improvement of 0.18.The experimental results show that the algorithm is not influenced by the differences caused by limited sensors.It still has certain universality for different data sources.In addition,the algorithm has significantly reduced the amount of operations and the time efficiency is increased by 85%.It has high usability and practical application value.