基于无人机正射影像的河道水域动态监测模型的研究——以浙江省宁波市鄞州区为例

Orthophoto image of unmanned aerial vehicle-based study ondynamic monitoring model of river water area——A casestudy of Yinzhou District of Ningbo City in Zhejiang Province

利用卫星影像开展河道水域非法占用监测,存在后期数据处理周期长、实时性弱等不足。为此,基于无人机正射影像图提出一种集Mask RCNN网络与图像差值计算的河道水域动态监测模型,通过机器识别改善现存问题。首先对正射影像进行拼接、矫正与分割,获取原始数据集后利用有监督学习方式训练Mask RCNN模型。最后获取河道轮廓图并计算两期同一位置正射影像的差值与阈值对比,来判定河道水域是否发生变化。试验结果表明,在机器识别与人工处理对比中本方法具有平均90%左右的准确率。本模型准确性较高,可辅助作为小型河道水域非法占用的实时动态监测工具。

The monitoring of the illegal occupation of river water area with satellite images has the defects of long post-processing cycle of data,weak real-time performance,etc.Therefore,a model for the dynamic monitoring of river water area integrated with Mask RCNN network and image difference-value calculation is proposed based on the orthophoto image of unmanned aerial vehicle,so as to improve the existing problems through the machine recognition.At first,the Mask RCNN model is trained with the mode of supervised learning after the acquisition of the original data through the stitching,correction and segmentation of the orthophoto images.Finally,the outline of river is obtained and then the difference-value between the orthophoto images in two stages is calculated and compared with the relevant threshold value for judging whether river water area is changed.The experiment result shows that the mean accuracy of this method is about 90%in the comparison between machine recognition and manual processing,thus it can be assistantly used as a tool for real-time dynamic monitoring of illegal occupation of small river water area.