摘要
目前对土地利用变化检测的研究在多时相遥感影像数据上进行,而在第三次全国国土调查(简称“三调”)工作中,通过套合矢量图斑数据与高分辨率遥感影像并完全依赖目视解译提取不一致图斑,该方法耗时,且仅抽取少量数据进行提取正确性质量检查。使用卷积神经网络模型,采用在ImageNet数据集中预训练的MobileNetV2模型对结合了SIRI-WHU数据集的贾汪区分割后数据集进行训练,将分类结果与矢量图斑数据进行比对,提取出不一致图斑,以检验该模型在不一致图斑正确性检查中的可行性。实验证明,通过该方法能够提取出不一致图斑,有望在“三调”质量检查以及土地利用变化检测等工作中推广。
Nowadays land use change detection researches are conducted based on the multi-temporal remote sensing images.During the work of the Third National Land Survey,the inconsistent spans are extracted based on the high-resolution remote sensing images linked with the latest land survey database,relying entirely on visual interpretation.This method is time-consuming and only a small amount of data is extracted to check the accuracy and quality of extraction.We adopt convolutional neural network(CNN)model to solve this problem.MobileNetV2 model pretrained by ImageNet database is adopted to train the fragmented dataset of Jiawang area combined with SIRI-WHU datasets.Then,we compare the classification results with vector spans data to extract inconsistent spans,to prove the feasibility of the model in detecting the correctness of extracted inconsistent spans.The experiment proves that the inconsistent spans can be extracted by the proposed method.It is expected to be applied in quality inspection of the Third National Land Survey and land use change detection.
作者
庞馨妍
刘伟
PANG Xinyan;LIU Wei(School of Geography,Nanjing Normal University,Nanjing 210023,China;School of Geography,Geomatics and Planning,Jiangsu Normal University,Xuzhou 221116,China;State Key Laboratory of Resources and Environment Information System,Beijing 100101,China)
出处
《测绘地理信息》
CSCD
2022年第S01期125-130,共6页
Journal of Geomatics
基金
国家自然科学基金(42071362)
江苏省自然科学基金(BK20191373)
关键词
空间数据质量检查
不一致图斑
卷积神经网络
迁移学习
土地利用分类
spatial data quality audit
inconsistent spans
convolutional neural network(CNN)
transfer learning
land use classification
