Apple Orchard Recognition And Area Extraction Based On Multi-Temporal Remote Sensing And GEE Platform

China is an important producer and exporter of apples.There are more than 80 countries producing apples in the world.China’s apple planting area and output account for more than50% of the world.Apple planting areas in China are mainly concentrated in northern regions,such as Shandong,Hebei,Shaanxi,Shanxi,Gansu,etc.The apple industry has an important impact on promoting China’s economic development.In Shandong and other major planting provinces,apple planting has become a pillar industry of rural revitalization,agricultural insurance,and government subsidies.Therefore,it is of great significance to accurately identify the location of apple orchards and extract their area.Based on the Google Earth Engine cloud platform and Sentinel-2 remote sensing images of three time phases,this study took the apple orchard in Qixia City,Shandong Province as the research area,and determined the best time phase for apple orchard recognition after analyzing the spectral curves and vegetation index curves of six land types,including apple orchard,other orchards,cultivated land,forest and grassland,construction land and water area;Based on the best time phase,the importance of the recognition features of apple orchard is analyzed,redundant features are eliminated,the optimal features are selected,different features are combined to develop the apple orchard recognition scheme,and random forest classifier is used to evaluate the accuracy of various schemes to select the best feature combination scheme;On the basis of the best feature combination scheme,random forest classifier,support vector machine classifier,decision tree classifier and Bayesian classifier are used to identify the apple orchard,and the classifier with the highest accuracy is selected to extract the area of the apple orchard.The main results of this study are:(1)Determine the best time phase of apple orchard recognitionBased on the spectral characteristics of field measured points on Sentinel-2 images from apple orchards,other orchards,cultivated land,forest and grassland,construction land,and water areas,the curve characteristics of six land types from March to October were analyzed,and the vegetation index curves of six land types were calculated.The results showed that the curve characteristics of apple orchards were significantly different from those of other land types during apple germination,flowering,and fruit ripening periods.Therefore,during apple germination,flowering The fruit ripening period is the optimal phase for apple orchard recognition,and a multi temporal data set for apple orchard recognition is constructed.(2)The best feature combination for apple orchard recognition was screenedBased on multi-temporal data sets,random forest classifier is used to analyze the importance of spectral features of apple orchard recognition,add features one by one,select the first 11 spectral features most suitable for apple orchard recognition according to the change of overall accuracy,and then combine texture features,vegetation index features,optimal spectral features,and all spectral features into six different feature combination schemes for apple orchard recognition,The results show that the "texture feature+vegetation index feature+optimal spectral feature" scheme has the highest accuracy,which is 4.34% to17.89% higher than the other five schemes.(3)The best classification method for apple orchard recognition is obtained and the area is extractedOn the basis of the best feature combination,the random forest classifier,support vector machine classifier,decision tree classifier and Bayesian classifier are used to identify the apple orchard.The overall accuracy is 93.75%,90.91%,87.81% and 82.62% respectively.The KAPPA coefficients are 0.92,0.89,0.85 and 0.79 respectively.The accuracy of the random forest classifier is 2.84% to 11.13% higher than that of the other three classifiers,so the recognition accuracy of the random forest is the highest,The effect of apple orchard recognition is the best.The area of Qixia apple orchard extracted based on random forest classifier is 651.68km2,accounting for 32.32% of the total area of Qixia.At the same time,the analysis of the spatial distribution of apple orchards shows that the apple orchards in Qixia City are mainly concentrated in Yangji Town,Guandao Town,Guanli Town,Shewopo Town,Sikou Town and Xicheng Town in the southwest.The area of apple orchards accounts for more than 40% of the area of towns,while the distribution in the middle and east is relatively small.Based on the Google Earth Engine platform,this study used the multi-temporal Sentinel-2 image to accurately identify and extract the area of Qixia apple orchard,providing a reference for fruit tree identification in other regions,and providing data support for the development of apple industry in this region.