Research On Vegetation Recognition Method Based On High Resolution Remote Sensing Data

Correctly grasping the spatial distribution characteristics of surface vegetation is the basic work of agricultural resource measurement.At present,remote sensing images are the main information source for statistics and monitoring of ground crops.With the continuous improvement of remote sensing technology,the information contained in them is more abundant.How to efficiently extract information from high-resolution remote sensing images has become the key to the application of remote sensing technology.In remote sensing image research,vegetation recognition based on traditional pattern recognition and classification algorithms is more common.This method requires manual selection of recognition features and settings.The application of parameters is time-consuming and labor-intensive and the accuracy is not high,and it is gradually difficult to meet the needs of remote sensing applications.Nowadays,convolutional neural networks have developed rapidly in the field of image processing,and have achieved excellent results in target detection,image segmentation,and scene interpretation.In the field of high-resolution remote sensing image processing,image target recognition based on convolutional neural networks is one of the current research hotspots.In the past research,it is found that there is less research on large-scale regions and multi-source data fusion.At the same time,in the face of the diversity of remote sensing images,there are few feature extraction models designed for targets in specific remote sensing images.Based on the above problems,this article aims to automate vegetation feature extraction and species recognition.Through the separate processing and research of satellite remote sensing and unmanned remote sensing data,convolutional neural network is the main research method,and the remote sensing image data processing method,Improve and innovate in the construction of vegetation recognition model and analysis of experimental results.A series of specific research work and related results are as follows:(1)Aiming at the multi-spectral image from satellite,which has the characteristics of multi-scale features in the image and large amount of information in the image itself,this paper constructs an improved model based on FCN to realize the vegetation identification method of satellite remote sensing image,by using hole convolution The method reduces the depth of the model and the amount of parameters,and improves the speed of model training.For the problem of the scale difference of the target features in satellite images,the multi-scale convolution method is used to increase the diversity of the model’s receptive field,so that the model can "pay attention" to vegetation targets of various scales.Finally,a training test was conducted on the satellite remote sensing data set for identifying ginseng vegetation areas established in this paper,and the test confirmed the effectiveness of the improved method in this chapter.(2)For UAV remote sensing images with high spatial resolution,due to the limitation of sensors,the data in the image only contains RGB spectral information,which has a serious impact on accurate vegetation identification.Starting from information,through remote sensing image fusion,new images with rich spectral information are obtained.Aiming at the characteristics of the new image,a C-Net network model is proposed,which can better identify honeysuckle,corn and soybeans in the fused remote sensing image.By comparing the original UAV image and the fused image,the effectiveness of feature information enhancement is proved.Compared with traditional vegetation recognition methods such as artificial neural networks,support vector machines,and nearest neighbor algorithms,it is verified that this method effectively improves the results of UAV image vegetation recognition.