Self-supervised learning-based oil spill detection of hyperspectral images

作     者：DUAN PuHong XIE ZhuoJun KANG XuDong LI ShuTao DUAN PuHong;XIE ZhuoJun;KANG XuDong;LI ShuTao

作者机构：College of Electrical and Information EngineeringHunan UniversityChangsha 410082China Key Laboratory of Visual Perception and Artificial Intelligence of Hunan ProvinceChangsha 410082China 

出 版 物：《Science China(Technological Sciences)》 (中国科学（技术科学英文版）)

年 卷 期：2022年第65卷第4期

页      面：793-801页

核心收录：

学科分类：0810[工学-信息与通信工程] 083002[工学-环境工程] 0830[工学-环境科学与工程（可授工学、理学、农学学位）] 07[理学] 08[工学] 081002[工学-信号与信息处理] 0713[理学-生态学] 

基　　金：supported by the National Natural Science Foundation of China (Grant No. 61890962 and 61871179) the Scientific Research Project of Hunan Education Department (Grant No. 19B105) the Natural Science Foundation of Hunan Province (Grant Nos. 2019JJ50036 and 2020GK2038) the National Key Research and Development Project (Grant No. 2021YFA0715203) the Hunan Provincial Natural Science Foundation for Distinguished Young Scholars (Grant No. 2021JJ022) the Huxiang Young Talents Science and Technology Innovation Program (Grant No. 2020RC3013) 

主　　题：hyperspectral image self-supervised learning data augmentation oil spill detection contrastive loss 

摘      要：Oil spill monitoring in remote sensing field has become a very popular technology to detect the spatial distribution of polluted ***,previous studies mainly focus on the supervised detection technologies,which requires a large number of high-quality training *** solve this problem,we propose a self-supervised learning method to learn the deep neural network from unlabelled hyperspectral data for oil spill detection,which consists of three parts:data augmentation,unsupervised deep feature learning,and oil spill detection ***,the original image is augmented with spectral and spatial transformation to improve robustness of the self-supervised ***,the deep neural networks are trained on the augmented data without label information to produce the high-level semantic ***,the pre-trained parameters are transferred to establish a neural network classifier to obtain the detection result,where a contrastive loss is developed to fine-tune the learned parameters so as to improve the generalization ability of the proposed *** performed on ten oil spill datasets reveal that the proposed method obtains promising detection performance with respect to other state-of-the-art hyperspectral detection approaches.