基于卷积神经网络的无人机影像违章建筑检测应用

无人机航拍影像具有分辨率高、回访周期短等特点，利用无人机遥感技术手段对城市范围的建设进行动态监测，可及时、有效地发现涉嫌违法的建设活动。本文结合实际项目需求，研究通过卷积神经网络方法进行违章建筑的自动检测，替代过去靠大量人力检查的模式，目前测试区域无人机影像试验取得了较好的效果，在样本数据不足5000份的情况下，准确率和召回率分别达到了71%和88%。随着样本数据的不断增多，基于该深度学习方法将较大程度上持续提升检测准确率和召回率，能够更精准地发现违法活动，具有较大的实际应用价值及潜力。

The application of illegal building detection from VHR UAV remote sensing images based on convolutional neural network

UAV aerial photography has the characteristics of very high resolution(VHR) and short revisit period. By using UAV remote sensing technology to dynamically monitor the construction activities in urban areas, suspected illegal construction activities can be identified promptly. In this paper, the method of detecting and discovering illegal building by using convolutional neural networks in some project on remote sensing data production is researched. Thus, past mode of manual inspection can be replaced. Good results are achieved in the test areas where the sample data is less than 5000 with the optimal precision of 71% and recall of 88%. With the continuous increase of sample data, the precision and recall rate can be improved greatly based on the proposed method, and illegal activities can be found more accurately. The research shows great potential applications.