Classification Of Complex Orchards And Segmentation Of Densely Planted Fruit Trees Using Multi-platform Remote Sensing Data

China is the world’s largest producer and consumer of apples,and the planting area and output of apples exceed 50% of the world’s total output.In Shandong,Shanxi,Shaanxi,and other advantageous apple production areas,the apple industry has become a pillar industry of economic development.In order to realize the precise management and monitoring of apple orchards,it is very important to accurately obtain the distribution of orchards and the canopy structure of fruit trees.The development of remote sensing technology has promoted the generation of high-resolution and high-time-domain remote sensing images.Through the processing and information extraction of remote sensing images,the means of monitoring the distribution and growth of apple orchards will be enriched,and the production of apple production areas in my country will be further guaranteed.and sustainable development.In this paper,two main apple producing areas,namely Linyi County,Shanxi Province and Qixia City,Shandong Province,are used as research areas,and the distribution of apple orchards and the information of apple tree canopy are extracted by using satellite remote sensing data and UAV remote sensing data respectively.Orchard distribution information is of great significance to the government’s macro-control,optimization of planting layout,and efficient agricultural production;single fruit tree canopy information provides important information for the realization of smart agriculture and differentiated planting management.In general,the work of this paper is divided into two parts.The first part uses time-series multispectral satellite images to extract the phenological characteristics of different fruit trees.to extract.The second part is the high spatial resolution image of the apple orchard obtained by the digital camera of the drone,and the image segmentation watershed algorithm is used to segment the single tree and extract the fruit tree canopy of the apple orchard in different periods.The main contents and achievements of this paper are as follows:(1)In the study area,due to the variety of fruit trees,the complex growth environment,the characteristics of fragmentation and narrowness,and the growth phenological cycles of different types of fruit trees are basically the same,which provides a better classification for fruit trees.big challenge.The traditional DTW(Dynamic Time Warping,DTW)algorithm can only further distinguish the category by calculating the difference of a single curve,so it does not have an advantage in the classification of fruit trees with relatively consistent phenological curves.In this study,an entropy-weighted DTW model based on optical and SAR(Synthetic Aperture Radar,SAR)data integration is proposed.Using the exponential time series of optics and SAR,the DTW distance of each curve is calculated,and the entropy weight method is integrated to classify the classification.The five index features with the largest contribution are finally used to perform fine fruit tree classification using the minimum distance of the curve.The results show that at the block scale,the overall classification accuracy of the method proposed in this paper is 0.75,and the Kappa coefficient is 0.68,which is significantly improved compared with the single-band DTW classification results.Finally,in order to facilitate the better generalization of the proposed model,we explore the classification performance of the model in pixel-level data,the overall classification accuracy is 0.63,and the Kappa coefficient is 0.51,which is lower than the classification results at the parcel scale.In the overall results of the classification,the pixel-level results are consistent with the plot-level results,which provides a theoretical basis for the future promotion of the model.This method explores the distribution of orchards with high similarity,and the obtained orchard distributions can provide favorable information for the government to optimize the planting structure and control the macroeconomic benefits of the fruit industry.(2)Fruit tree canopy information is the basis for monitoring fruit tree growth and diagnosing health status.In this paper,single-tree canopy extraction was carried out in the period of vigorous fruit tree growth and high canopy closure.The fruit tree canopy in the orchard has the characteristics of complex canopy texture,regular arrangement,similar crown width,and no prominent local commanding heights.It is difficult to achieve single tree segmentation.The traditional watershed algorithm based on seed point marking is not ideal for fruit tree segmentation.Therefore,this paper designs a single tree segmentation method based on UAV digital images.It uses Gaussian filtering,morphological operations,and adaptive threshold segmentation to extract potential canopy subjects,and uses the potential canopy subjects as regional seed blocks.The watershed algorithm implements single wood segmentation.In order to verify the robustness of the algorithm,this paper conducts experiments using digital images of three densely planted apple orchards.A total of 1025 fruit trees in the study area were counted and single-tree canopy extraction was performed.The accuracy results showed that the overall precision rate of tree number statistics was 99.09%,the recall rate was 95.22%,the F1 score was 97.11%,and the overall single-tree canopy extraction accuracy rate is 93.45%.Then,the algorithm proposed in this paper is compared with the single-point seed point single-tree segmentation algorithm.The results show that the under-segmentation error between fruit trees based on the regional seed block method is low,and the overlapping area of the tree canopy can be effectively suppressed.A fruit tree,which can preserve the canopy outline of the fruit tree to the greatest extent.Finally,this paper adds a segmentation experiment of the algorithm in the digital surface model and Ex G(Excess Green Index,Ex G)images.The results show that the regional seed block is also suitable for Ex G images derived from UAV digital images,and achieves better results.segmentation result.This method has made exploratory work for the extraction of single fruit tree canopy in densely planted orchards,and provided important information for the precise implementation of water and fertilizer management to improve the intelligent management of orchards.