摘要
针对大范围快速监管黄河中游采砂情况以维护其生命健康的问题,提出了一种加权平均集成UNet算法和PSPNet算法,改进损失函数,并结合遥感影像特点合理设定参数的集成学习算法。通过利用算法训练得到的采砂信息提取模型对实地调查的5个采砂点进行了信息提取,结果显示:UNet算法提取的找全率(Recall)为79.87%,准确率(Precision)为15.80%,交并比(IoU)为16.75%;PSPNet算法提取的找全率(Recall)为57.57%,准确率(Precision)为27.79%,交并比(IoU)为31.17%;集成学习算法提取的找全率(Recall)为89.57%,准确率(Precision)为55.72%,交并比(IoU)为60.28%。因此,本文算法可以在一定程度上应用于河湖两侧采砂信息的提取,更好地辅助水利行业强监管的执行。
Aiming at the large range quick supervision of the sand mining in the Mid-Yellow River for maintaining the life and health of the river, an ensemble learning algorithm that integrates the UNet algorithm and PSPNet algorithm by weighted average, improves the loss function and reasonably sets the parameters in combination with the characteristics of the remote sensing images is put forward herein. Through the sand-mining extraction model obtained by the relevant algorithm training, the information extractions of five sand-mining points for the in situ investigation are carried out, from which the results show that the recall rate of 79.87%, the precision rate of 15.80% and the IoU of 16.75% are extracted by the UNet algorithm and the recall rate of 57.57%, the precision rate of 27.79% and the IoU of 31.17% are extracted by the PSPNet algorithm, while the recall rate of 89.57%, the precision rate of 55.72% and the IoU of 60.28% are obtained from the ensemble learning algorithm. Therefore, the algorithm proposed herein can be applied to the extraction of the sand-mining information on both the sides of river and lake for better assisting the implementation of strong supervision within the water sector.
作者
王守志
奚歌
张福坤
刘金玉
耿振云
詹昊
张云姣
WANG Shouzhi;XI Ge;ZHANG Fukun;LIU Jinyu;GENG Zhenyun;ZHAN Hao;ZHANG Yunjiao(China Water Resources Beifang Investigation,Design and Research Co.,Ltd.,Tianjin 300222,China)
出处
《水利水电技术》
北大核心
2020年第12期161-168,共8页
Water Resources and Hydropower Engineering
基金
河湖监管卫星遥感地图智能比对技术需求研究(JGZXSY2019-26)。
作者简介
王守志(1989-),男,工程师,硕士,主要从事遥感影像数据在人工智能处理方面的应用研究。E-mail:Httpwmz@163.com。
