本文已被:浏览 305次 下载 676次
投稿时间:2023-05-15 修订日期:2023-05-31
投稿时间:2023-05-15 修订日期:2023-05-31
中文摘要: 为满足输电线路山火易发地区的低漏检、高精度、大范围、高时效性火点近实时监测需求,本文以地球同步轨道卫星影像为例,提出一种基于MC-CNN的山火检测算法。该算法其主要特点包括:通过取OTSU算法与上下文算法两类算法的并集来进一步增加潜在火点的可能性,进而在一定程度上降低火点检测的漏检率;引入PCA算法对输入特征进行优化,构建多通道网络结构以及基于联合概率和PSO参数寻优算法获取不同通道火点识别权重,在加权平均的基础上最终判定火点;以固定高温热源和太阳耀斑作为虚假火点进行去除,降低误报率。本文随机选取2019年至2022年期间输电线路附近历史卫星监测山火案例,利用已知火点样本对火点反演结果进行验证所提算法的有效性,计算获取火点检测精度为89.4%。
Abstract:To meet the needs of low residual, high precision, large range, high timeliness wildfire point near the real-time monitoring for transmission line fire prone areas, this paper put forward a fire detection algorithm based on MC - CNN.with the earth synchronous orbit satellite images.The main characteristic of this algorithm are as follows:By taking OTSU algorithm with the context of two kinds of algorithm union set to further increase the possibility of potential point, which to a certain extent, reduce the residual rate of fire detection;Get weights of different channel fire point recognition by optimizing the input features with the PCA algorithm, building a multichannel network structure and optimization algorithm based on joint probability and PSO parameters ,and determine the fire on the basis of the weighted average finally.This paper randomly selected the satellite monitoring history fire case near transmission line from 2019 to 2022,and inverse analysis the effectiveness of the proposed algorithm by using samples of known fire points,results show that the calculation for the fire detection accuracy is 89.4%.
keywords: geostationary satellite wildfires monitoring multi-channel convolutional neural network joint probability weighted average false fire spot removal
文章编号: 中图分类号: 文献标志码:
基金项目:
作者 | 单位 | |
王国芳 | 云南电网有限责任公司电力科学研究院 | 1045238377@qq.com |
文刚* | 云南电网有限责任公司电力科学研究院 | 1192381484@qq.com |
马仪 | 云南电网有限责任公司电力科学研究院 | |
周仿荣 | 云南电网有限责任公司电力科学研究院 | |
王一帆 | 云南电网有限责任公司电力科学研究院 | |
马御棠 | 云南电网有限责任公司电力科学研究院 | |
耿浩 | 云南电网有限责任公司电力科学研究院 |
引用文本: