Research On Key Methods For Building Information Acquisition Of Hollow Village Based On High-resolution Image

In recent years,with the rapid development of urbanization,the rural population and capital in the suburban areas have gradually shifted from rural areas to cities,resulting in a large number of idle and abandoned buildings in rural areas,and hollowization phenomenon has gradually evolved into a larger area in rural settlements as Hollow village.The emergence of Hollow Village has caused a serious waste of land resources,which has restricted the improvement of the rural living environment and hindered the sustainable development of the rural economy.The research on the renovation of Hollow Village is one of the key issues for the promotion of new rural construction and the building of a well-off society in China's urban and rural transitional period.The use of UAV remote sensing technology for rapid acquisition of Hollow Village building information has a positive effect on the scientific renovation of Hollow Village.In this paper,the outdoor collection of Hollow Village buildings are vulnerable to the weather,easy to miss measurement,low efficiency of indoor visual interpretation,and heavy workload.Therefore,we introduce the deep learning methods and UAV remote sensing technology to study the key technologies of Hollow Village building information acquisition.Firstly,the preprocessing of UAV high resolution remote sensing imagery(Unmanned Aerial Vehicle High-resolution Remote Sensing Imagery)was studied,which provided a basic image for the detection and information acquisition of hollow village buildings;Secondly,a variety of building interpretation models are analyzed,and selected convolutional neural network(CNN)and migration learning methods to construct building interpretation model.Then,based on the interpretation model constructed by the convolutional neural network combined with the saliency analysis method,the detection of the buildings in the hollow village of small scale was realized.The interpretation model constructed based on the migration learning method combined with the multi-scale segmentation technology realized the detection of the buildings in the hollow villages in a large area.Finally,using the building detection method proposed in this paper,combined with UAV remote sensing and mobile GIS technology construct a Hollow Village building information survey system.The research work and achievements of this paper mainly include the following aspects:1.Aiming at the problem that the precision of the POS(unmanned aerial vehicle)data is not high enough,the original image outside orientation element is not accurate enough and difficult to estimate,a non-iterative algorithm based on linear transformation is proposed to solve the outer azimuth of the UAV.This method can solve higher-accuracy outer azimuth elements without the least square iteration,and realizes the correction of POS data.In addition,aiming at the problem of the error of the reference control point in the registration of traditional remote sensing images,a method of UAV image registration based on weighted global least squares(WTLS)is proposed.The polynomial regression coefficients are estimated by introducing WTLS to improve the accuracy of image registration.2.To solve the problem of low efficiency and high visual interpretation of traditional visual interpretation,this paper constructs a hollow village building sample database based on UAV high score images,and introduces a convolutional neural network and migration learning method to construct a hollow village building solution with Translation model.Experiments show that the classification accuracy of the interpreting model constructed in this paper can reach more than 90%,and it can achieve accurate and efficient interpretation of hollow village buildings.3.A hollow building detection method based on convolutional neural network is proposed.First of all,using multi-scale saliency detection to obtain the significance area containing building information,obtain the main objective and weaken the interference effects of other unrelated objects,reducing the data redundancy.Then,through the sliding window,the target sample blocks in the saliency region are obtained,and these sample blocks are input into the trained interpreting model for classification.The detection of the buildings in the small-scale hollow village is realized.Experiments show that the overall accuracy rate can reach 81%.4.For the problem of insufficient samples in complex scenes,a migration learning mechanism was introduced to propose a detection method that is applicable to a large range of hollow village buildings.Firstly,the object-oriented segmentation technique is used to segment the UAV high-resolution image to obtain several meaningful image regions.Then,the segmentation regions are classified according to the translation model and the sliding window constructed by the migration learning,and the results are marked and implemented a wide range of buildings.Finally,combining the mobile GIS technology,UAV remote sensing technology and the building detection method proposed in this paper,a preliminary integration of a hollow village building information survey system is achieved,which enables the rapid collection and management of hollow village building information.