Dynamic Change Of Forest Types Based On Multi-source Remote Sensing Data

Forests are an important part of the global biosphere.The complex and diverse internal structure makes the process of forest ecological changes very obvious.Therefore,real-time understanding and grasping of forest type distribution and its dynamic change information has important practical significance.Most of the forest classification research at this stage is based on a single remote sensing data or a single classification model.The lack of multi-algorithm comparison based on multi-source remote sensing data cannot guarantee the optimal classification accuracy,which greatly restricts the subsequent development of dynamic changes.In response to this problem,this article selects Jingzhou Miao and Dong Autonomous County in Huaihua City,Hunan Province as the study area,using ZY-1-02C images,Planet images and GF-1 images as data sources,combined with ground control points and Jingzhou county boundary vector files Using four classification algorithms:Maximum Likelihood Method,BP Neural Network Method,Support Vector Machine Method,and Random Forest Method,to classify the forest types of the preprocessed multi-source remote sensing data,and use UAV images and Google Earth The image verification data evaluates the accuracy of the classification results of different algorithms.The problem of the best classification algorithm based on multi-source remote sensing data is studied,and the application potential of high-resolution remote sensing images in forest type classification is fully explored.In addition,this article further analyzes the 2017-based on the optimal classification model(random forest method).The results of forest type classification in Jingzhou County in 2019 were post-processed,and the classification data of 2018 and 2019 with good classification results were selected,and the forest type dynamic change study was carried out in Gantang Town,which is rich in forest types in the northeastern part of Jingzhou County.,Has accumulated experience for the application of high-resolution remote sensing satellite data in the detection of forest type dynamic changes.The main research results include:(1)The random forest method has a good classification effect:the land types in the study area are divided into four types:broad-leaved forest,coniferous forest,bamboo forest and non-forest land,using maximum likelihood method,BP neural network method,and support vector The four classification algorithms of machine method and random forest method were used to classify the forest types in the study area by remote sensing.The results show that the random forest method has the best classification effect.The overall classification accuracy in the ZY-1-02C image,Planet image and GF-1 image are 86.87%,88.24%,and 90.37%respectively;the Kappa coefficient is 0.83 respectively.,0.84,0.87;the classification effect of the maximum likelihood method is the second,the overall accuracy is 84.53%,86.98%,87.45%;the Kappa coefficient is 0.79,0.82,0.83 respectively;the classification effect of the support vector machine method is poor,and the overall accuracy They were 68.04%,70.27%,79.42%;Kappa coefficients were 0.57,0.60,0.72;BP neural network method had the worst classification effect,with overall accuracy of 53.54%,81.83%,and 73.57%;Kappa coefficients were 0.37,0.76,0.64.It shows that in the classification of high-resolution remote sensing images,the random forest method has great advantages.(2)GF-1 image has a good classification effect:Random forest method is used to extract forest type information from ZY-1-02C image,Planet image and GF-1 image in the study area,and the result shows the classification effect of GF-1 image The best,the overall classification accuracy is 90.37%,the Kappa coefficient is 0.87;the classification effect of Planet image is the second,the overall classification accuracy is 88.24%,the Kappa coefficient is 0.84;the classification effect of the ZY-1-02C image is poor,and the overall classification accuracy It is 86.87%,and the Kappa coefficient is 0.83.It shows that GF-1 images have great application potential in the extraction of forest type information from high-resolution remote sensing images.(3)Analysis of dynamic changes of forest types:Select the results of the classification results of forest types in Jingzhou County in 201 8 and 2019 with good classification results,and conduct a study on the dynamic changes of forest types in Gantang Town in the northeastern part of Jingzhou County,which is rich in forest types..From 2018 to 2019,the overall change of forest types in Gantang Town was relatively gentle,and the changes among forest types tended to be stable:bamboo forests changed the most,and the total area showed a downward trend.The total area was reduced to 2.43km2,a decrease of 33.38%;The total area of the leaf forest has changed greatly,and the total area is increasing.The total area is increased to 1.73km2,and the increase rate is 5.80%;the coniferous forest has a small change,and the total area is increasing.The total area is increased to 0.63km2,and the increase rate is 1.15.%;The total area of non-forest land shows a slight increase trend,with a total increase of 0.07km2,an increase of 0.15%.It shows that Gantang Town,Jingzhou County,is in the process of regional development of townships from 2018 to 2019.Green ecological agriculture and township greening have been developed in concert,the effect of township landscape has been improved,the living environment of residents has improved,and the construction of ecological civilization and social and economic development have been achieved.However,there are still some forest logging,farmland reclamation,and road construction in the forest area.