Research And Application Of Intelligent Recognition Method Of Forest Land Type From Remote Sensing Images

Natural resources survey is a survey of the common characteristics of natural resources.Its main task is to find out the distribution and scope of various natural resources on the surface,as well as their development status in human activities.By analyzing this information,it can be used to analyze the natural resources of the whole country.Have a basic understanding of the background status and common characteristics of resources.In the process of each survey,the latest remote sensing technology in the current period will be combined with the icing on the cake.The current stage is the stage of rapid development of deep learning,and the application of deep learning in remote sensing image recognition is also a current hot spot.The existing deep learning image recognition is mostly carried out in natural scenes,the targets are relatively concentrated,and the shapes and sizes are relatively regular;while the woodland that needs to be recognized in this article is carried out on remote sensing images,and the targets are of different sizes and shapes.And the lack of corresponding training data sets poses a great challenge to the accuracy of the recognition results.Based on deep learning,this paper proposes to apply it to remote sensing image recognition of forest land types to provide a new method for improving the recognition efficiency.It starts from three parts:data set production,intelligent recognition model and result transformation theory research.First,analyze the characteristics of forest land type remote sensing image data,design the data set file organization according to the common deep learning data set format,and propose a forest land remote sensing image data set production method based on the analysis results,and use QGIS and Python to implement this method,completed the link of data set production,which provided data support for model training;then built an intelligent recognition model of woodland type remote sensing images based on Mask R-CNN,trained the model,and finally got the training weight file,input the test data,after feature extraction,candidate area generation,pooling,etc.,in the fully connected layer classification and regression,as well as segmentation mask prediction,and finally get the recognition results,verify the generalization ability of the model,and also design a verification experiment to use accuracy,recall,and AP value on the verification set,a comprehensive evaluation and analysis of the model’s recognition ability on forest land remote sensing images;finally,a theoretical study of converting the raster into a vector and generating an attribute table based on the recognition results.The experimental results show that the method of making data set based on QGIS used in this article can efficiently complete the task of data set production;and the forest land type remote sensing image intelligent recognition model based on Mask R-CNN can realize the instance segmentation task of forest land type remote sensing image.The recognition results provide a reference for the staff,and the AP value can reach up to 0.833;the final recognition result conversion theory research is carried out to realize the editing of the recognition results,and provides a method basis for the subsequent realization of the editing of the recognition results.