Research On Fine Crops Classification Of Hyperspectral Remote Sensing Crops Based On Conditional Random Field

Agriculture is the foundation of the country’s development.The accurate identification of crop classes is an important link in the process of agricultural modernization.It plays an important role in the application of agricultural disaster monitoring,crop yield estimation,growth analysis,determination of crop category area and spatial distribution.At the same time,it is also an important basis for the rational allocation of resources,scientific adjustment of agricultural structure,and planning of economic development strategies in the agricultural production process.UAVs is a kind of unmanned aerial vehicle,which can be controlled by a computer preset program or a remote control device.It has become an important means of ground observation and provides support for the development of precision agriculture.By mounting a hyperspectral sensor on the drone,you can obtain remote sensing data with both high temporal resolution,high spatial resolution,and high spectral resolution,thereby using richer and more subtle spectral and spatial information,and more accurate agricultural conditions information.But at the same time,it also causes the phenomenon of " different objects with the same spectrum" and " the same objects with different spectrum ",so it is important to use this information to the maximum extent.The classification method based on probability graph model can make full use of contextual information with the help of graph theory knowledge.Random field model is a kind of probability map model,and it is one of the research hotspots based on spatial context information classification method.Markov random field is a typical random field model that can model the context information of classification marks,but it requires that the observation information is independently distributed under the conditions of a given mark,which limits the flexibility of spatial information utilization.The conditional random field model overcomes the shortcomings of the Markov random field model,reduces the requirement for probability distribution,directly models the posterior probability of a given observation data,and can consider context information in both the label data and the observation data.So it has modeling flexibility.Therefore,this study takes the farmland of Honghu City and Hanchuan City,Hubei Province,China as the research object,and uses the UAV hyperspectral remote sensing platform to obtain high spatial resolution hyperspectral remote sensing images as experimental data.In the framework of the conditional random field model,the fusion of image spectral information,spatial context information,spatial feature information,and spatial location information is carried out to carry out fine crop classification research based on UAV hyperspectral remote sensing.This article mainly researches and implements the following:(1)Fusion texture information based on conditional random fields for the fine classification of crops in UAV-borne hyperspectral remote sensing imagery.Aiming at the problem that traditional crop classification methods are limited to the use of spectral information,a suitable potential function is designed in the framework of a conditional random field model to fuse texture information based on the overall perspective and spatial context information based on the pixel perspective.While using texture information for classification,it also effectively alleviates the problem of excessive smoothing in traditional conditional random field models.(2)Spatial–spectral fusion based on conditional random fields for the fine classification of crops in UAV-borne hyperspectral remote sensing imagery.According to the characteristics of crop planting distribution and type,on the basis of extracting spectral information and texture information,multi-angle spatial features such as end elements and morphology are fused,and the multi-angle spatial-spectral fusion features of high spatial resolution hyperspectral images are modeled in the framework of conditional random field models to more fully dig out the potential information of images.(3)Fine crop classification using spectral-spatial-location fusion based on conditional random fields for UAV-borne hyperspectral remote sensing imagery.Modeling the spatial position information of high spatial resolution hyperspectral remote sensing images by considering the interaction between a large range of pixels,and mining large-scale spatial interaction information.In the framework of the conditional random field model,a potential function that combines spectral information,multi-angle spatial feature information,local spatial context information,and spatial position information is constructed to improve the precision of fine crop classification.