光谱特征和空间卷积相协同的近岸海域养殖塘遥感信息提取  

Information extraction of coastal aquaculture ponds based on spectral features and spatial convolution

作　　者：李叶繁 王琳[1,2,3] 张冬珠 LI Yefan;WANG Lin;ZHANG Dongzhu(College of Environmental and Safety Engineering,Fuzhou University,Fuzhou 350108,China;Academy of Geography and Ecological Environment,Fuzhou University,Fuzhou 350108,China;Fujian Provincial Key Laboratory of Remote Sensing of Soil Erosion and Disaster Prevention,Fuzhou 350108,China)

机构地区：[1]福州大学环境与安全工程学院,福州350108 [2]福州大学地理与生态环境研究院,福州350108 [3]福建省水土流失遥感监测评估与灾害防治重点实验室,福州350108

出　　处：《自然资源遥感》2022年第4期42-52,共11页Remote Sensing for Natural Resources

基　　金：国家自然科学基金重点支持项目(促进海峡两岸科技合作联合基金)“闽台高植被覆盖区台风暴雨型滑坡孕灾环境、成灾机制与动态预警”(编号:U2005205)资助。

摘　　要：为控制水产养殖塘无序发展带来的负面效应,促进水产养殖业进一步发展,首要解决的就是对其快速、准确识别和提取的问题。水产养殖塘是被复杂道路和堤坝分割的特殊网状水体,单纯的光谱特征或空间纹理特征都不足以对其准确提取,且混合特征规则集对计算机性能要求越发苛刻。鉴于此,以Landsat影像序列为数据源,基于谷歌地球引擎(Google Earth Engine,GEE)平台,提出了一种结合影像光谱信息、空间特征和形态学操作的沿海水产养殖塘自动提取方法。该方法联用了双特征水体光谱指数(改进型组合水体指数(modified combined index for water identification,MCIWI)与改进的归一化差异水体指数(modified normalized difference water index,MNDWI))以突出大面积水体与养殖塘的网格特征,再利用低频滤波空间卷积运算拉伸养殖与非养殖水体之间的差异特征,将水产养殖塘区作为一个整体准确识别和快速提取。研究结果表明:①该方法总精度达到93%,Kappa系数为0.86,典型区域叠加比对检验流程验证,提取结果和实际结果重叠比例均在90%以上,平均重叠比例达92.5%,反映了提取方法的高精度和可靠性;②2020年福建省近岸海域水产养殖塘区总面积为511.73 km^(2),主要分布在漳州市、福州市和宁德市;③核密度分析结果表明漳州市的水产养殖塘集聚度高,相应其养殖塘管理压力也较大。该方法可以实现近岸海域水产养殖塘的自动化提取,对促进渔业养殖的有序管理和科学发展具有重要的意义。To control the negative effects of the disorderly development of aquaculture ponds and promote the further development of the aquaculture industry,the top priority is to realize rapid and accurate identification and extraction of information on aquaculture ponds.Aquaculture ponds are special net-like water bodies divided by complex roads and dikes.Simple spectral features or spatial texture features are not enough for accurate information extraction.Moreover,the mixed feature rule set gets more demanding on computer performance.Therefore,based on the Landsat image sequence and the Google Earth Engine(GEE)platform,this study proposed an automatic extraction method for coastal aquaculture ponds,which combined the image spectral information,spatial features,and morphological operation.In this method,dual characteristic water spectral indices,that is,the modified combined index for water identification(MCIWI)and the modified normalized difference water index(MNDWI),were employed to highlight the grid characteristics of large water bodies and aquaculture ponds.Then,the low-frequency filtering spatial convolution operation was used to stretch the differences between aquaculture and non-aquaculture water bodies.Finally,the information on aquaculture pond areas as a whole was identified and extracted accurately and quickly.The results are as follows.①This method has overall precision of 93% and a Kappa coefficient of 0.86.According to the test process verification of typical area superposition comparison,the overlapping proportions between the extraction results and the actual results were all more than 90%,averaging 92.5%,reflecting the high precision and reliability of this extraction method.②In 2020,the coastal aquaculture ponds in Fujian Province had a total area of 511.73 km^(2) and were mainly distributed in Zhangzhou,Fuzhou,and Ningde cities.③The kernel density analysis suggested that Zhangzhou had a high concentration of aquaculture ponds and thus had high pressure in the management of aquaculture ponds.Thi

关 键 词：水产养殖塘区 阈值分割 空间卷积 GEE平台 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]