摘要
红树林对维护生物多样性以及生态平衡等具有重要意义。因此,高效、精确地提取红树林植被信息以及实时对其进行监测十分必要。本文提出了一种高分辨率遥感影像红树林像素级精确提取的深度学习方法。针对红树林遥感分类精度不高的问题,通过强化图像中心信息,弱化边缘信息的方法引入CLoss损失函数,添加Dropout、Batch Normalization层构建了适用于红树林识别的CU-Net模型,采用滑动重叠拼接方法构建了新的预测模型,有效解决了预测结果边缘信息不足以及有拼接痕迹的问题。将本文方法的识别结果与U-Net、SegNet、DenseNet模型的预测结果以及传统的SVM、RF方法进行对比,结果表明,本文模型相较于其他深度学习模型泛化能力更强,识别效果更好,在两个测试区域的平均总体精度、平均交并比分别达到了94.43%、88.12%,平均F1-分数在红树林和普通树木的精度分别达到了95.96%、90.49%,精度明显高于传统的SVM、RF方法,也高于其他几种神经网络方法,验证了该模型在红树林识别领域的有效性,可为高分辨率遥感红树林识别领域提供一条新的思路。
Mangroves are important for maintaining biodiversity as well as ecological balance.Therefore,it is necessary to extract mangrove vegetation information efficiently and accurately and to monitor it in real time.A deep learning method for pixel-level accurate extraction of mangroves from high-resolution remote sensing images is presented in this paper.For the problem of low accuracy of mangrove remote sensing classification,CU-Net model for mangrove identification is constructed by introducing CLoss loss function by strengthening image center information and weakening edge information,and adding Dropout and Batch Normalization layers.And a new prediction model is constructed by sliding overlap splicing method,which effectively solves the problem of insufficient edge information and splicing traces in the prediction results.The recognition results of the proposed method are compared with the prediction results of U-Net,SegNet and DenseNet models as well as the traditional SVM and RF methods.The results show that the proposed model has stronger generalization ability and better recognition effect compared with other deep learning models.In the two test areas,the average OA and MIoU reach 94.43% and 88.12%,respectively.The average F1-score in mangrove and ordinary trees reach 95.96% and 90.49%,respectively.The accuracy is significantly higher than that of traditional SVM and RF methods,as well as several other neural networks.The effectiveness of the model in the field of mangrove recognition is verified,which can provide a new idea for the field of high resolution remote sensing mangrove recognition.
作者
蔚铭阁
芮小平
邹亚荣
张茜
Yu Mingge;Rui Xiaoping;Zou Yarong;Zhang Xi(School of Earth Sciences and Engineering,Hohai University,Nanjing 211100,China;National Satellite Ocean Application Service,Ministry of Natural Resources,Beijing 100081,China;Key Laboratory of Space Ocean Remote Sensing and Application,Ministry of Nat-ural Resources,Beijing 100081,China)
出处
《海洋学报》
CAS
CSCD
北大核心
2023年第3期125-135,共11页
基金
海南省重点研发计划(高新技术专项)
国家自然科学基金(41771478)。
关键词
深度学习
红树林识别
高分辨率遥感影像
卷积神经网络
deep learning
mangrove identification
high-resolution remote-sensing images
convolutional neural network
