Research On Vegetation Ecological Changes Using UAV Remote Sensing Combining Spectral Data And Leaf Area Index

Quantitative analysis of vegetation ecological changes in the Lijiang River Basin through the use of remote sensing images can provide scientific evidence and quantitative reference for ecological environment protection in the region.In this study,a representative area of the Lijiang River Basin with typical landforms was selected as the research object,and high-resolution multispectral images were obtained using unmanned aerial vehicles(UAVs).Combined with deep learning models,a quantitative analysis model for surface vegetation classification was established to achieve monitoring of typical vegetation ecological changes in the Lijiang River Basin.The main research content is as follows:(1)Construction of a drone multispectral vegetation extraction model based on deep learningThe first key point of this research is to use deep learning methods to construct a drone multispectral vegetation extraction model.In order to train the convolutional neural network model,it is necessary to capture and pixel-level annotate multispectral images to establish a dataset,ensuring the reliability and accuracy of model training and testing.The large amount of pixel-level annotation work is challenging,and the complex topography of high-resolution images makes annotation difficult.This paper proposes a solution,using traditional machine learning classification based on spectral features to pre-classify images,and then visually interpret and modify them to solve the problem of data scarcity.This paper then discusses the issue of using deep learning to segment vegetation in multispectral images and proposes a 3D-UNet model for multispectral image segmentation.This model takes into account both spectral and spatial context information.The model is compared with different types of datasets to analyze its performance.The overall Io U of the 3D-UNet is 3.8% higher than the UNet,and the model parameters have been reduced by 98.4%.The results show that the model can provide accurate prediction segmentation performance.(2)Inversion prediction and change analysis of Leaf Area Index(LAI)The second focus of this paper is to predict and analyze the changes in the Leaf Area Index(LAI),which is an important indicator of plant growth.This paper introduces an innovative "segment first,then invert" strategy to invert and predict the global leaf area index based on high-resolution multispectral images obtained by drones.Using field-sampled leaf area index and multiple vegetation spectral indices calculated based on multispectral images,linear and exponential inversion regression models are established,the best model is fitted,and the global leaf area index is predicted.Finally,the LAI level and overall vegetation coverage level of two periods of images in the study area are compared.The LAI has decreased by 53.14%,12.64%,34.20%,72.08%,94.10%,74.01% respectively in 6 grading levels,and the vegetation coverage has decreased from43.498% to 24.492%.This achieves monitoring of vegetation ecological changes in the study area and provides valuable quantitative data for regional ecological environmental protection.