Tree Detection System Based On Multi-source Remote Sensing Data

High-resolution satellite remote sensing images and LIDAR point cloud data are very widely used in forestry applications.Both of these data can be used to obtain single tree information.High-resolution satellite remote sensing images are easier to obtain,and the location information and canopy area information of the trees can be detected in these data,but the height information is lacking.LIDAR point cloud data can be detected in the tree location information,canopy area information and height information,but the cost of such data collection is higher.The detection of single tree information in remote sensing data faces the following problems: 1,In different forestry applications,often need to use different remote sensing data to detect single tree information;2,Some single tree detection methods are parameters dependency and the results' accuracy need to be improved.In order to solve the above two problems,this thesis has developed a single tree detection technology based on multi-source remote sensing data.The main contents of this thesis are as follows:1: In order to extract the tree information in the high-resolution satellite remote sensing image,this thesis presents a single tree detection method based on the deep neural network.First,in order to reduce the parameter dependency of the method and generalize the performance,this thesis uses the BP neural network method to detect the single tree.In order to further improve the recognition accuracy of the method and reduce the commission rate and the omission rate of the method,the five characteristics of energy,entropy,mean value,skewness and kurtosis are selected and added to the BP neural network model to form a deep neural network model.In this thesis,six experimental regions were selected as the test area.The experimental results show that the deep neural network method has good detection effect.2: In order to extract the tree information in the LIDAR point cloud data,this thesis presents a single tree detection method based on gradient direction clustering.First,the point cloud data is rasterized to generate the digital surface model of the experimental area.The gradient orientation is used to cluster the digital surface model to obtain the candidate set of single tree.The noise is removed by a mathematical morphology-based corrosion and expansion operation.Then select a single tree based on the shape parameters and density parameters.The results show that the method is more excellent than the other detection algorithms in the single tree detection dataset of NEWFOR.3: Based on the above two kinds of single tree detection method,this thesis designs a single tree detection system.The system can handle high-resolution remote sensing image data and LIDAR point cloud data,which can detect tree location information,tree canopy size information and tree height information from the data.