| In the process of natural resource management and ecological civilization construction along the key shoreline of the Yangtze River Economic Belt,traditional visual interpretation is difficult to meet the requirements of dynamic,efficient and flexible shoreline land-use/land-cover(LULC)change monitoring.The application of high precision,multi-spectral and multi-temporal remote sensing image with computer automatic interpretation technology in shoreline monitoring can effectively improve the efficiency of LULC extraction.Taking Jianyang section of Tuojiang River shoreline as the study area,this paper applies the object-oriented method to extract the LULC features based on GF-2 satellite imagery,and combines qualitative and quantitative methods to explore the optimal parameters for multi-scale segmentation of various surface features.Secondly,different classification methods are comprehensively compared in terms of accuracy and efficiency based on the results of single-level segmentation,and the CART Decision Tree method with the best comprehensive effect is selected for pruning and weighted optimization.The optimized CART Decision Tree is combining with the multilevel image segmentation structure to extract the LULC information.Finally,the characteristics of LULC change and landscape pattern change in the study area from 2016 to 2019 were analyzed based on the two phases of LULC data.The research method is applicable to areas with similar features along the Tuojiang River shoreline,and is able to provide a reference for the dynamic change monitoring of shoreline areas in China.The conclusions are as follows:(1)The optimal segmentation parameters(scale parameters,shape factors and compactness factors)of multi-level and multi-scale segmentation are explored by the use of trial and error method,maximum area method,the mean variance method and the Estimation Scale Parameter 2(ESP2)tool verification method.The image of study area is divided into three layers,and the segmentation parameters are [160,0.1,0.4];[110,0.4,0.6];[60,0.3,0.4].(2)Combining the results of feature selection,single-level segmentation is performed on the experimental images.the same samples and features are used to extract LULC information by four classification methods: Cart Decision Tree(CART),Random Forest(RF),Support Vector Machine(SVM)and K-nearest Neighbor(KNN).Performances on classification accuracy and efficiency of each method are compared.The results show that CART classification method is better than the other three classification methods..(3)The multi-level CART decision tree classification model is formed by constructing a multi-level classification system and combining it with the optimized decision tree rule set,which can not only make different objects belong to the corresponding segmentation scale,but also automatically determine the land category description rule set.By applying the multi-level classification system to LULC classification of the two phases of the shoreline,the results show that: the overall classification accuracy reaches 0.863,with the kappa coefficient 0.819 and the classification time 145 seconds.The classification accuracy and efficiency are both better than single-level methods.(4)The change of LULC types was relatively stable,and the overall pattern of cultivated land and woodland remained unchanged in the study area from 2016 to 2019.However,the conversion among some LULC types was active,mainly including the conversion from cultivated land to forest and urban expansion.During the three years,the area of cultivated land decreased by 9.71%,and the area of cultivated land with a slope greater than 25° decreased by more than 40%.woodland occupied the most in high-slope land,and the woodland area of the two phases accounted for more than 70% of the land with a slope above 25° and showed an increasing trend.The area of urban,industrial,and mining land increased by 3.28%,of which the growth points were mainly located in flat areas and were relatively intensive.The transportation land has developed rapidly,with an increase of 34.47% in the past three years,forming links between regions.In the meantime,the changes of other land types were relatively stable.(5)The overall change of LULC landscape pattern in the study area was obvious,with increasing land use intensity and deepening degree of landscape fragmentation,while the patch shape is developing towards the direction of artificial and regularization,and the patch distribution tends to be uniform.The landscape diversity was rich,and the LULC types tend to be constant.The ecological balance was preserved well,and various landscape types were intertwined and interdependent. |
PDF Full Text Request