一种基于机器学习的火点检测算法  

A fire detection algorithm based on machine learning

作　　者：黄宇飞[1] 徐嘉[1] 李智慧[2] 张菁[2] HUANG Yu fei;XU Jia;LI Zhihui;ZHANG Jing(Beijing Institute of Spacecraft System Engineering,Beijing 100000,China;School of Computer Science and Technology,Harbin Engineering University,Harbin 150000,China)

机构地区：[1]北京空间飞行器总体设计部,北京100000 [2]哈尔滨工程大学计算机科学与技术学院,哈尔滨150000

出　　处：《测绘科学》2020年第10期64-70,共7页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(51679058);北京空间飞行器总体设计部研究项目。

摘　　要：针对传统基于多通道阈值火点判断方法在Landsat 8图像上选取困难的问题,该文提出一种基于机器学习的火点检测算法。该文采用非均衡数据分类框架,通过两个步骤实现火点检测:第一步为非火点排除,通过主分量分析提取特征,采用该文提出的正例优先感知机组合分类器排除掉大多数的非火点;第二步为精确分类,通过线性判别分析变换得到特征,采用加权支持向量机实现准确的火点判别。实验结果表明,该文算法实现了准确率与漏检率的较好折中。A fire detection algorithm based on machine learning was proposed in this paper to improve traditional methods of multi-channel thresholds which are difficult to determine in Landsat 8 images. The classification frame of imbalanced data was adopted and fire detection was implemented by two steps. The first step was to exclude non-fire points. The features were extracted by principal component analysis transformation. Most non-fire points were eliminated by the positive sample priority perceptron group classifier proposed in this paper. The second step was to accurately classify the remaining samples. The features were calculated by linear discriminant analysis. Weighted support vector machine was used to implement precise classification. The experiment results showed that the algorithm had better trade-off between accuracy and missed detection rate.

关 键 词：火点检测 正例优先感知机 线性判别分析 加权支持向量机 

分 类 号：P237[天文地球—摄影测量与遥感]