| Rural buildings are one of the basic materials for observing rural land changes and economic development.The combination of deep learning algorithms and drone images to extract rural buildings can solve the problems of time-consuming and labor-intensive manual statistics in rural areas,as well as the lack of building data.It is of great significance to provide data support for rural resource planning and smart agriculture construction.In response to the difficulties in extracting feature information caused by the complex terrain,diverse size of buildings,and irregular layout distribution in rural China,this article selects four deep learning algorithms: U-Net algorithm,Seg-Net algorithm,Deeplabv3+,and PSPNet algorithm for recognition.Based on the comprehensive consideration of experimental results,U-Net algorithm is selected for improvement.Through comparative analysis of optimization mechanisms,the U-Net algorithm was innovatively integrated into the coordinate attention mechanism,and the results showed that the overall effect of the integrated coordinate attention mechanism was outstanding.The main contents of this study are as follows:(1)Dataset production.The data set used in this study was prepared by preprocessing and expanding 1100 aerial drone images from some rural areas in China.(2)Application of deep learning algorithm in building recognition.U-Net algorithm,Seg-Net algorithm,Deep Labv3+ algorithm and PSPNet algorithm were selected to perform recognition and extraction experiments on buildings in UAV images.After comparing these three models through experiments,and considering comprehensively,the U-Net algorithm with better recognition effect and shorter time consumption is selected for improvement.(3)Improvement of U-Net algorithm.By comparing and analyzing the optimization mechanism,the innovative fusion of U-Net algorithm and coordinate attention mechanism for building recognition has improved the accuracy by 3%,with significant improvement effects.(4)Independently designed UAV image building identification system development.Combining the improved U-Net algorithm with the Flask framework and Vue framework for practical applications aims to enable convenient,fast,and simple building identification and display. |
