摘要
在强光的照射下,水体的镜面反射往往会对遥感影像产生很大影响,其主要表现就是在图像上产生大小不同、形状各异的亮斑。这些亮斑附近的地物信息基本上都被淹没,对后期的影像分析会造成不同程度的影响,因此对这些亮斑的检测识别就显得尤为重要。文章以DeeplabV3plus为主要网络,提出一种融合Swin-Transformer模块的网络模型。该模型将Swin-Transformer网络作为一个模块与卷积骨干网络并行提取特征。提取出的两类特征经上采样后进行特征融合,再经多次卷积等实现了水体亮斑的识别与分割。实验结果表明,该模型能够对不同类型、不同形状的水体亮斑进行识别分割,其平均交并比为93.44%。
The specular reflection of water bodies often has a significant impact on remote sensing photos when exposed to high light,which is primarily seen as brilliant spots of various sizes and shapes.The detection and identification of these bright spots are particularly crucial because the feature information of ground objects near these bright spots is essentially drowned,which affects the subsequent image analysis to varying degrees.Taking DeeplabV3plus as the backbone network,this paper proposes a network model that integrates the Swin-Transformer module.The model extracts features from the Swin-Transformer network as a module in parallel with the convolutional backbone network.The extracted two types of features are up-sampled for feature fusion,and then the recognition and segmentation of the bright spots of water bodies are realized by multiple convolutions,etc.The experimental results show that the model proposed in this paper can recognize and segment different types and shapes of water bright spots,and its average intersection and the merging ratio is 93.44%.
作者
陈毅夫
何敬
刘刚
毛佳琪
CHEN Yifu;HE Jing;LIU Gang;MAO Jiaqi(College of Earth Sciences,Chengdu University of Technology,Chengdu 610059,China;State Key Laboratory of Geohazard Prevention and Geoenvironment Protection,Chengdu University of Technology,Chengdu 610059,China)
出处
《遥感信息》
CSCD
北大核心
2023年第4期129-136,共8页
Remote Sensing Information
基金
成都市技术创新研发项目(2022-YF05-01090-SN)
成都理工大学研究生质量工程项目(2022YJG022)
地质灾害防治与地质环境保护国家重点实验室项目(SKLGP2018Z010)
四川省科技计划项目(2021YFG0365)
四川省自然资源厅项目(kj-2021-3)。
