Spatial Information Extraction Method Of Mountain Banana Forest Based On Gf-6 And Sentinel-2remote Sensing Image

Bananas are the main cash crops,and China's banana acreage ranks among the highest in the world.There have been a large number of researches at home and abroad that use remote sensing technology to classify crops and extract information,but there are few remote sensing studies on banana forests or banana forests on plateaus and mountains.Yunnan is located on the Plateau,the mountain bananas are the main crop of Yunnan and characteristic plateau crops in the province.The development of plateau-specific agriculture has promoted the development of farmers and the local economy,and has also accelerated the process of agricultural modernization.Accurately obtaining the plantation information and spatial distribution information of the banana forest laid the foundation for the ecological suitability evaluation and yield evaluation of the planting environment.As agricultural production activities are generally have unstable factors such as large planting area,large regional planting differences,and inconsistent economic benefits per unit area,and Yunnan have large plateau and mountainous area,it has brought unfavorable conditions to traditional surface farmer surveys.It is the traditional survey methods that can't fit with the needs of today's production,the improvement of remote sensing satellite image resolution and the rapid development of remote sensing technology have brought new solutions to this problem.This paper selects Jinping County,the main banana producing county in Yunnan,and uses GF-6 data and sentinel data to construct multi-source data.Based on the traditional identification of land types,based on the Google Earth Engine cloud platform integrated classifier,object-oriented extraction methods.The information extraction of banana forest in Jinping County was studied,and the effects of different data sources and different classifiers on the information extraction effect and expression level of banana forest were compared,there is the result:(1)Research on Banana Forest Information Extraction Based on GF-6.Because the banana forest has unique texture features that can be used as a large recognition feature,used the object-oriented classification method.After repeated comparisons and optimizations,the segmentation scale 40 and the scale parameter 80 was selected as theoptimal segmentation scale.Texture information,DEM elevation information,and vegetation index were combined to construct multi-source data.Because the Random Forest classification has the advantages ofprocessing large data,less manual intervention,and inhibiting overfitting,random forest classification is used to extract information from banana forests in the study area.The accuracy of the extracted banana forest users was 97.37%,the mapping accuracy was 96.2%,the misclassification error was 2.63,the missed error was 3.8,the KAPPA coefficient was 0.96,and the overall classification accuracy was 97.35%.The extraction of banana forest information based on the GF-6 has the best extraction effect and can represent the distribution of banana forests.However,due to the high resolution of the image,there are also problems such as slow calculation and low efficiency.(2)Research on Banana Forest Information Extraction Based on Sentinel-2.The Google Earth Engine cloud platform's Sentinel-2 image was used to classify it using the platform's integrated support vector machine(SVM),classification decision tree(CART),and random forest.Comparing the results of three different classification methods,because the support vector machine has the advantage of being able to obtain the main information from multiple features,the support vector machine have best extraction effect in the three methods are used to extract the banana forest information from sentinel-2 images.Preferably,the distribution area of the banana forest is consistent with the data obtained from the field survey.The user precision of the obtained banana forest is 93.9,the mapping accuracy is 93.3,the misclassification error is 6.1,the missed error is 6.7,the KAPPA coefficient is 0.93,and the overall classification accuracy is94.66%.(3)Based on the research,The methods can roughly represent the banana forest information.By comparing the extraction effects of the above methods,the plantation area of banana forests extracted based on the random forest classification of GF-6 is closer to the statistical yearbook,and the distribution locations are closer to the planting locations of field investigations.The most accurate extraction of banana forest information based on sentinel data is the result obtained by support vector machineclassification.It can be seen that the optimal classification method is different for different images.The higher the resolution of the image under the same classification method,the better the classification accuracy can be effectively improved.