Research On The Identification Of Tea Plantations Based On Multi-Source Remote Sensing Data And Its Recognition Accuracy Scale Effect

As the origin of tea in the world,my country has a wide distribution of tea planting areas,which are mainly divided into arbor tea and terrace tea.Arbor tea has a good ecological environment and rich biodiversity,while terrace tea plantations are mostly planted in a row at the expense of natural forests,arable land and fertile land.Largescale and uncontrolled planting has brought many negative effects on the ecological environment,such as the reduction of land cover types,the destruction of species diversity,etc.,which has brought many irreversible ecological and environmental problems.Yunnan province is one of China's major tea producing regions and has seen a significant expansion of tea plantations in recent years.Therefore,the dynamic monitoring of tea gardens has important ecological and environmental protection significance,and can provide reference for decision-making and ecological restoration of relevant departments.However,tea gardens mostly planted in tropical and subtropical mountainous areas have broken spatial distribution and are mostly surrounded by orchards,shrubs and other vegetation.Therefore,it is a huge challenge to identify tea gardens,it is difficult to extract tea gardens from the complex environment only with spectral features.However,tea plantations are different from other evergreen natural forests,shrubs and deciduous rubber forests in their special planting methods and management models.Regular pruning makes tea plantations form a special artificial phenological period,and the rows of cultivation methods form a special the spatial texture structure,which provides an important basis for remote sensing identification.This paper takes Dayi tea plantations in Menghai County,Yunnan Province as the research area,and constructs multi-spatial scale data based on multiple remote sensing images to extract the phenological features and spatial texture features of tea plantations for identification of tea plantations respectively,and determines the best spatial scale for identifying tea plantations based on phenological features and the best spatial scale and best time window for identifying tea plantations based on texture features.The main research conclusions are as follows:(1)The phenological features extracted based on time series data can extract tea plantations with high precision,and the accuracy is affected by the spatial resolution,where 10 m recognition accuracy is the lowest and 15 m recognition accuracy is the highest.Phenological features were extracted from time series data with different spatial resolutions of 10 m,15 m,20 m,30 m and 60 m to identify tea gardens,The 10 m recognition accuracy is the lowest,The 15 m recognition accuracy is the highest,the overall accuracy is 94.17%,the Kappa coefficient is 0.86,the producer accuracy is95.28%,and the user accuracy is 85.82%.Compared with the 10 m recognition result,the overall accuracy has increased by 2.69%,the Kappa coefficient has increased by0.07,and the producer accuracy and user accuracy have increased by 10.24% and0.78%,respectively.(2)The tea plantation can be accurately identified by texture features,and the accuracy increases first and then decreases with the decrease of spatial resolution,with the lowest accuracy at 0.5 m and the highest accuracy when the spatial resolution decreases to 14 m and 15 m.Based on the remote sensing images of observation at multiple spatial scales(0.5m,2 m,10 m,15 m and 20 m)extract texture features to identify tea plantations,the accuracy increases first and then decreases with the decrease of spatial resolution.When the resolution is 0.5 m,the recognition accuracy is the lowest,the overall accuracy is 83.18%,and the Kappa coefficient is 0.58.When the resolution is 15 m,the recognition accuracy is the highest,the overall accuracy is 93.27%,the Kappa coefficient is 0.84,the producer accuracy is 88.98%,and the user accuracy is 87.60%.Compared with the 0.5 m recognition result,the overall accuracy has increased by10.09%,the Kappa coefficient has increased by 0.26,and the producer accuracy and user accuracy have increased by 21.26% and 15.93%,respectively.Based on the series of spatial resolutions(0.5 m,2 m,4 m,6 m,8 m,10 m,12 m,14 m,16 m,18 m,20 m)remote sensing images obtained by the nearest neighbor scale conversion method to extract texture features to identify tea plantations,the overall accuracy and Kappa coefficient showed a trend of first increasing and then decreasing as the spatial resolution decreased.When the resolution is 0.5 m,the recognition accuracy is the lowest,when the resolution is 14 m,the recognition accuracy is the highest,the overall accuracy is 91.03%,and the Kappa coefficient is 0.78,which is7.85% and 0.2 higher than the recognition accuracy of 0.5 m respectively.(3)When using texture features to identify tea plantations,the budding period is the best identification time window,and the identification accuracy is the highest.To analyze the best time window when using texture features of visible remote sensing images for tea plantation recognition,three images with different time phases of tea plantation before pruning,after pruning and budding period were selected to extract texture features for tea plantation recognition,among which the accuracy of tea plantation recognition before pruning is lowest,in the budding period(early March),the recognition accuracy is the highest,the overall accuracy is 93.27%,the Kappa coefficient is 0.84,the producer accuracy is 88.98%,and the user accuracy is 87.60%.Compared with the recognition accuracy before pruning,the overall accuracy is increased by 11.43%,the Kappa coefficient is increased by 0.27,the producer accuracy and user accuracy are increased by 14.18% and 21.63%,respectively.