| Forest is the most extensive terrestrial ecosystem,accounting for about one-third of the total land area.It is also an essential part of the earth's ecosystem.Whether the forest ecosystem is normal is closely related to regional ecological security and sustainable development of social economy.Traditional artificial-based forest resource survey methods have disadvantages such as high cost,high intensity of field work,and long time-consuming,and their timeliness and accuracy often fail to meet the standards of practical applications.Traditional single data source is difficult to obtain high-precision forest parameter information,and multi-source data fusion will become an effective method for forest resource monitoring in the future.The 13th five year plan of national forestry development proposes to strengthen the construction of ecological detection and evaluation system,deepen the comprehensive application of remote sensing,positioning and communication technology,and build an integrated monitoring and early warning evaluation system of sky and ground.With the continuous popularization of intelligent forestry,the continuous progress of information technology and earth observation technology,the use of different types of remote sensing images,UAVs and wireless communication network to monitor the information of forest resources can effectively guide forestry production.Through the construction of a comprehensive monitoring system of forest resources,we can accurately understand the basic situation and changes of forest resources in real time,Timely response measures can greatly improve the effect of resource protection and utilization,which is of great significance for environmental governance and ecological construction.From the perspective of forest resource monitoring integrated with space and earth,and aiming at the problems of complex and diverse forest resource monitoring information types,high data dimensions,high information redundancy,and large amount of data information processing,this paper studies a comprehensive and three-dimensional forest resource monitoring network system,researches network solutions and models suitable for forest environment monitoring wireless sensor network,and focuses on solving the problem of self-organization and multi-hop transmission efficiency of sensors in a network structure that conforms to the model of the forest environment monitoring wireless sensor network,and researches the integrated solutions of massive remote sensing image data processing,data storage and data mining,carries out research on the comprehensive monitoring system of forest resources to provide theoretical support and data reference for the optimization and reform of the continuous forest resource inventory system.The main work of this paper is as follows:(1)Aiming at the data transmission problem of wireless sensor network for forest environment monitoring,an improved multicast routing model based on private network is proposed.This model is particularly suitable for determining the relative position of large-scale and low-density wireless sensor nodes,and it is also suitable for network monitoring environments with poor connectivity between nodes and large ranging errors between long-distance nodes.On the basis of studying the topological structure of neighboring nodes in wireless sensor networks,combining mainstream shape learning and nonlinear dimensionality algorithms,a local combined localization(LCL)algorithm is proposed.This algorithm is based on the paired distance between each node and neighboring nodes within a certain communication range,obtain the local geometric information of the global structure by constructing a local subspace for each node first,and then compare these subspaces to get the internal global coordinates of all nodes,and finally combine the global structure and anchor node information,use the least square algorithm to calculate the absolute coordinates of all unknown nodes.And obtain comprehensive forest environmental information on weather,light,soil and air quality by constructing a monitoring system in the Harbin Experimental Forest Farm area.(2)For the preprocessing and data mining of remote sensing data,we use SequenceFile,a binary file storage form provided inside Hadoop to serialize image data into a byte stream and store it in a binary file.When the MapReduce task is executed,it is directly read by Hadoop's SequenceFilelnput Format(input file format),which realizes parallel image processing.By improving the remote sensing image feature parallel extraction algorithm through a custom partition strategy,and increasing the number of Reduce tasks in the MapReduce program for extracting remote sensing image features,the parallelization of Hadoop's remote sensing image feature extraction has been realized,so as to improve the efficiency of parallel processing.Aiming at the problem that K-Means clustering algorithm needs to manually determine the initial clustering center and the number of clusters,so that the clustering results fall into the local optimum,this paper combines the Canopy algorithm to improve the K-Means algorithm.First,use the Canopy algorithm to "roughly cluster" the feature information of the remote sensing image,and then use the result as the initial clustering point of the K-Means clustering algorithm to complete the remote sensing image classification processing and provide high-quality remote sensing images for the monitoring data.(3)Aiming at the problem of low detection accuracy existing in the existing remote sensing image change detection model,a two-stage remote sensing image change detection model is proposed.This model makes full use of the multi-dimensional features of remote sensing images,uses the U-net network to perform semantic segmentation of remote sensing images,and combines the classification results with the classification results of remote sensing data in different time phases,so as to accurately obtain the characteristics of the feature changes in the monitoring area.This method can effectively extract the texture and spectral features of remote sensing images and improve the accuracy of change detection.Using this model,the change detection of Xishuangbanna Nature Reserve based on convolutional neural network is realized.According to the application needs of vegetation change monitoring in the nature reserve,carry out system requirement analysis and design and model analysis,realize the forest monitoring system of Xishuangbanna Nature Reserve to accurate monitoring of forest changes in the region.In addition,this study explores the key technology of individual tree-level forest resources monitoring with high spatial resolution remote sensing images.The automatic individual tree crown delineation is conducted using H-maxima transform combined with mark control watershed algorithm,which makes the fine scale forest resources automatic monitoring possible. |
PDF Full Text Request