Knowledge-Based Land Cover Classification And Information Extraction In Coastal Zone Of Peninsular Malaysia

Landsat TM/OLI remote sensing image has the advantages of long time span and convenient access,and has become the main data source for large-scale land use information extraction.it is difficult to extract all the land types from the classification system with high accuracy only by the traditional computer automatic classification method when the spectral characteristics of the objects are similar or the types of land cover are complex.Therefore,for each type of land cover,it is of great theoretical and practical significance to fully mine Remote Sensing information,comprehensively use a variety of Geoscience knowledge,and establish a complete set of rules for extracting land use information with high efficiency and accuracy.Based on the existing classification system at home and abroad,a land cover classification system of coastal zone is established based on the current and regional characteristics of the land cover in the coastal zone of Malaysia Peninsula,including 8 first-class and 18 secondary.Combining with multi-geoscience knowledge and combining theory with experiment,appropriate feature vectors are selected to highlight the differences between different types of land cover and to construct corresponding rules for land classification extraction,the method of distinguishing land cover types by statistical analysis at pixel scale is improved.In the aspect of cultivated land extraction,based on multi-temporal images,using the difference of crop growth status between different cultivated land types in different periods,the cultivated land area is effectively extracted,and the distinction between paddy field and dry farm is realized.In the aspect of water extraction,five secondary types of ocean,shallow,lake,reservoir and river are distinguished by synthesizing characteristic information such as spectrum,elevation,geographical location and shape,and the problem that it is difficult to extract secondary types of water area only by using water body index is solved.In the aspect of vegetation types extraction,the differences of texture features,topographic distribution characteristics and spectral reflectance characteristics of different vegetation types are effectively utilized to distinguish natural forest,plantation forest,secondary forest,shrub and grassland.In the process of extracting construction land,the threshold value is set reasonably to divide the regional scopes of urban construction land and rural residential by using the nighttime lighting data,and the distinction between urban construction land and rural residential is realized through topological relationship,the problem that construction land and un-utilized land are difficult to distinguish because of the similar spectral features is sloved.The experimental results show that the rule set of classification based on multi-information such as temporal features,topographic features,texture features,shape features,topological features and spectral features has achieved good results in the extraction of secondary types of land cover.Its overall classification accuracy is more than 85%,and Kappa coefficient is higher than 0.8.Compared with the maximum likelihood supervised classification method,the classification accuracy is obviously improved,the overall classification accuracy is increased by 17%,and the kappa coefficient is increased by 0.17.Based on the knowledge rule set constructed in this paper,the land cover types of coastal zone of Malaysia Peninsula in 1990 and 2017 were extracted,and the differences of land cover changes in the eastern and western coastal zones of the Peninsula were analyzed from the overall and gradient aspects respectively.The results show that between 1990 and 2017,the reduction of natural forests in the western coastal zone was 31.65%,the reduction of natural forests in the eastern coastal zone was 28.12%.The overall development level of the western coastal zone is higher than that of the eastern coastal zone.In 2017,the proportion of development land in the western coastal zone reached 75.74%,while that in the eastern coastal zone was 52.54%.In addition,relying on the port,in the past 30 years,the development area of western coastal zone extends gradually to the island,and the development land in the area of 0-50 km away from the coastline shows a trend of gradual expansion.The main development area of the eastern coastal zone was concentrated in the range of 0-20 km from the coastline.By fully exploiting the multi-knowledge information besides the spectral features,the paper selects the appropriate feature indicators for different land types of coastal zones to establish the land cover classification rule set,and broadens the idea of land cover types extraction,which has important research significance for improving the accuracy of secondary classification of coastal land cover.In addition,Python and IDL language are used to compile the algorithm,which realized the automatic extraction of land cover types and greatly improved the work efficiency.At the same time,the knowledge rule set classification algorithm was applied to the extraction of land cover types in the eastern and western coastal zones of Peninsular Malaysia,the change and difference of land cover types in 1990 and 2017 were analyzed to provide important data support for the formulation of sustainable development policy for coastal zone of Peninsular Malaysia.