基于GAN的高光谱溢油检测实验方案设计  

Experiment Scheme Design of Hyperspectral Oil Spill Detection Based on Generative Adversarial Network

作　　者：宋冬梅[1] 王斌[1] 王海起[1] 樊彦国[1] SONG Dongmei;WANG Bin;WANG Haiqi;FAN Yanguo(College of Oceanography and Space Informatics,China University of Petroleum(East China),Qingdao 266580,Shandong,China)

机构地区：[1]中国石油大学(华东)海洋与空间信息学院,山东青岛266580

出　　处：《实验室研究与探索》2022年第12期5-8,73,共5页Research and Exploration In Laboratory

基　　金：国家自然科学基金项目(U1906217);山东省重点研发计划项目(2019GGX101033);山东省教学改革面上项目(M2021155);中国石油大学(华东)青年教师教学改革项目(QN201801);研究性教学方法改革项目(YK201802);专业建设与改革类项目(ZY202025)。

摘　　要：为实现高精度的海洋溢油高光谱遥感检测,设计了一种基于生成对抗网络(GAN)的高光谱溢油检测实验方案。方案利用光谱归一化和主成分分析方法对高光谱溢油图像进行预处理,以提高计算效率;构建以生成器和判别器对抗训练为特色的GAN网络模型,以实现端到端的溢油检测;利用训练好的判别器对溢油高光谱影像进行分类。为验证新方法的有效性,利用在2010年7月大连新港海上溢油事故中获取的机载高光谱数据开展了实验验证。结果表明,提出的方法在2套数据集上的总体分类精度相对于RF和SVM分别提升了2.33%和0.77%,Kappa系数达到了96.17%和95.56%。可见设计的方案能够实现高精度的溢油检测。To achieve high-precision detection of marine oil spills in the hyperspectral remote sensing field,an experimental scheme for oil spill detection based on Generative Adversarial Network(GAN)is designed.For improving the computational efficiency,firstly,hyperspectral oil spill images are pre-processed by using spectral normalization and Principal Component Analysis(PCA);secondly,a GAN model featuring the adversarial training between generator and discriminator is constructed to achieve end-to-end oil spill detection.Finally,the discriminator after training is used as a classifier to predict the hyperspectral oil spill images.To verify the effectiveness of the proposed method,the airborne hyperspectral data obtained in the offshore oil spill accident of Dalian Xingang in July 2010 are used to carry out experiments,and the experimental results show that the overall classification accuracy of the proposed method on the two datasets is improved by 2.33%and 0.77%compared with Random Forest(RF)and Support Vector Machine(SVM),respectively,and the Kappa coefficient reaches 96.17%and 95.56%,respectively.The above results show that the devised scheme enables high-precision oil spill detection.

关 键 词：溢油检测 高光谱图像 生成对抗网络 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]