Research And Application Of Grassland Classification Algorithm For Inner Mongolia Grassland Based On Multi-source Time-series Data Fusion

As the most extensive type of land cover in the world,grasslands are the largest natural resource on land in China and play an important role in ecological security and national economy.Inner Mongolia is rich in grassland resources and is an integral part of the ecological security barrier in the north of China.Grassland classification is a scientific approach to understanding grasslands and provides a theoretical basis for grassland production and practice.The current methods based on remote sensing images are suitable for small-scale grassland studies,while the grassland of Inner Mongolia covers a large area and the field sampling environment is harsh and difficult to investigate,so the classification study for the large scale and holistic grassland of Inner Mongolia is still a difficult task.Although the time series of vegetation indices can eliminate the problem of " same object different spectrum " and " foreign object same spectrum" based on a single remote sensing image,relying on the time series of a single vegetation index will lead to the underfitting of the model.Therefore,it is of great significance to fuse multi-source time series data for grassland classification to monitor the dynamics of grassland in Inner Mongolia.In this thesis,we investigate the classification of grassland types using multi-source time-series data for Inner Mongolia grassland meadows and design and implement a grassland classification system for Inner Mongolia grasslands,with the following main research works:(1)The Inner Mongolia grassland meadow dataset with multi-source time-series data fusion is constructed.In this thesis,we selected Inner Mongolia grassland resource data,combined with remote sensing time series data of growing seasons and climate time series data of similar years,and constructed a dataset with multi-source time series data fusion unique to Inner Mongolia grassland meadow,which provided data support for the proposed Inner Mongolia grassland meadow classification method and Inner Mongolia grassland meadow classification system,bridged the gap of multi-source time series data for grassland classification,and provided a reliable data source for the subsequent research of Inner Mongolia grassland.It provides a reliable source of data for subsequent research on Inner Mongolia grasslands.(2)A deep learning-based LSTM-Encoder-MF grassland classification method is proposed.A fusion network model combining LSTM,which is good at processing temporal data,and Transformer encoder,which has a multi-headed attention mechanism,is used to classify Inner Mongolian grassland meadow using multi-source temporal data.The experimental results prove that the classification model proposed in this thesis achieves93.75% accuracy and 94.03% F1 Score.(3)The grassland classification system of Inner Mongolia grassland was designed and implemented.The NDVI time series data and climate time series data were used to build a basic My SQL database to realize the unified management of the data.Geo Server was used to store remote sensing images and combined with Echarts visualization tool to realize the display of grassland-related data of Inner Mongolia grassland meadow in a large area and with high timeliness.The grassland classification system of Inner Mongolia grasslands was designed to realize large-scale monitoring of grassland changes in Inner Mongolia,to provide data support for the state and relevant government departments to formulate macro policies,to provide decision support for the benign use of grassland resources,and to play an important role in the ecological barrier of Inner Mongolia grasslands.