摘要
深度学习遥感信息提取技术在自然资源行业中在逐渐开展应用,但行业主要基于两期卫星遥感影像开展变化检测的方法进行情报图斑提取;广西具有典型亚热带沿海地域特点,多云多雨,光学遥感影像覆盖度较低,全域获取两期影像的周期较长;针对广西云上摄影时间周期相对较短和地貌破碎等特征,提出基于HRNet网络开展解译模型训练,建立符合广西地域特征的自然资源综合样本库,并通过采集获取单期的无人机遥感影像进行建筑,道路,水体,耕地,林地和园地进行智能分割提取,实现基于自主训练解译模型和已有遥感解译模型对高分辨率无人机遥感影像进行自动解译,具备提取速度快、准确率高的优势;经验证以上六类地物要素的准确率和召回率均超过85%;结合建设项目实施监管和耕地资源监测的工作实际,应用无人机智能解译成果,通过套合国土变更调查的耕地范围提取非农和非粮图斑,并套合建设项目审批图斑提取审批图斑范围外的建筑图斑,开展监测监管应用实践,并取得阶段性成效和探索。
Deep learning remote sensing information extraction technology is gradually applied in the natural resource industry.The information spots based on the detection method of two-satellite remote sensing image variation is extracted.There are the characteristics of typical subtropical coastal region in Guangxi province,which is cloudy and rainy,low coverage of optical remote sensing image is low,and long period of obtaining two images in the whole region.In view of the relatively short period of cloud photography and fragmentation of landforms,an interpretation model training based on the HRNet network is proposed to establish a comprehensive sample database of natural resources in line with the regional characteristics of Guangxi.The intelligent segmentation and extraction of buildings,roads,water bodies,cultivated land,woodlands and gardens are carried out by collecting and acquiring single-phase unmanned aerial vehicle(UAV)remote sensing images,which realizes the automatic interpretation of high-resolution UAV remote sensing images based on the autonomous training interpretation model and existing remote sensing interpretation model.It has the advantages of rapid extraction speed and high accuracy.The accuracy rate and recall rate of the above six types of ground feature elements are verified to be more than 85%.Combined with the supervision of actual construction project implementation and monitoring of cultivated land resources,the UAV intelligent interpretation results are applied to extract non-agricultural and non-grain patterns from the cultivated scope of the land change survey,and the construction patterns outside the construction project approval pattern scope are extracted from the construction pattern,which carries out the monitoring and supervision application practice,and it achieves the phased results and exploration.
作者
冯一军
陈霖
梁雄乾
王志虎
王鑫
FENG Yijun;CHEN Lin;LIANG Xiongqian;WANG Zhihu;WANG Xin(Guangxi Institute of Natural Resources Survey and Monitoring,Nanning 530023,China;Technology Innovation Center for Natural Resources Monitoring and Evaluation of Beibu Gulf Economic Zone,Ministry of Natural Resources,Nanning 530219,China;College of Computer and Information Engineeringof Guangxi Vocational Normal University,Nanning 530007,China;Beihai Campus of Guilin University of Electronic Technology,Beihai 541000,China)
出处
《计算机测量与控制》
2023年第6期231-237,共7页
Computer Measurement &Control
基金
广西重点研发计划(桂科AB22080077)
广西科技基地和人才专项(桂科AD20238044)
广西空间信息与测绘重点实验室基金(191851011)。
关键词
深度学习
人工智能
无人机
智能解译
监测监管
自然资源
deep learning
artificial intelligence
UAV
automatic interpretation
monitoring and management
natural resource
