自然资源多时相遥感影像智能匹配方法研究  

Multi-temporal Remote Sensing Images of Natural Resources Intelligent Matching Method

作　　者：史迪超 黎慧斌[2] 李亭谕 史晓明 SHI Dichao;LI Huibin;LI Tingyu

机构地区：[1]中国地质大学(武汉)地理与信息工程学院,湖北武汉430074 [2]广东省国土资源技术中心,广东广州510075 [3]湖北省航测遥感院,湖北武汉430074

出　　处：《地理空间信息》2021年第9期1-6,I0001,共7页Geospatial Information

摘　　要：遥感影像是全民所有自然资源资产清查工作的基础数据之一,时序影像匹配是多时相遥感数据服务于自然资源资产评估的关键。大量的自然资源资产清查单元(林地、耕地、湖泊、草地、湿地)位于地形平坦地区,采用传统影像匹配方法得到的正确匹配点较少,匹配精度较低。提出了基于视野感知的CNN-SIFT描述符,利用孪生网络模型感知同名点的视觉差异特征,根据地形自适应调节视野范围,从而增强匹配能力。结果表明,与邻域范围为128×128的SIFT描述符相比,CNN-SIFT描述符的正确匹配点平均增加了41.32%,正确匹配率提高了18.64%,湖泊水面、平原区块田等平坦区域匹配率提高了19.20%,能满足全民所有自然资源资产清查实际工作的要求,为其他卫星影像进行大范围大批量的正射影像生产提供借鉴。Remote sensing image is one of the basic data for natural resource assets inspection.And time series image matching is the key for multi-temporal remote sensing data to evaluate natural resource assets.A large number of natural resource assets(woodland,arable land,lake,grassland,wetland)are located in the terrain flat region,the gradient characteristics of whose are similar.There are few correct matching points obtained by traditional image matching method,and the matching accuracy is low.In this paper,we proposed CNN-SIFT descriptor based on visual field perception.We used the siamese network that was sensitive to visual differences to extract the visual field difference characteristics of the same name points,and adjusted the visual field adaptively according to the degree of flatness,which could enhance the matching ability and improve the matching accuracy.Experimental results show that comparing with SIFT descriptor with a neighborhood of 128×128,the matching points of CNN-SIFT descriptor increase more than 41.32%,the correct matching rate improves 18.64%,and the matching rates of complex waters and field improve 19.20%.This approach has been successfully applied to natural resource precision monitoring.The accuracy can meet the requirement of the actual work in the natural resource assets inspection,and this method can provide a reference for the large-scale orthophoto images production of other satellite images.

关 键 词：影像匹配 孪生网络 SIFT描述符 自然资源资产清查 CNN 

分 类 号：P237[天文地球—摄影测量与遥感]