基于随机森林的天气场景判别算法  

Weather scene recognition algorithm based on random forest

作　　者：史静[1] 朱虹[1] Shi Jing;Zhu Hong(The Faculty of Automation and Information Engineering, Xi＇an University of Technology, Xi＇an 710048, China)

机构地区：[1]西安理工大学自动化与信息工程学院,陕西西安710048

出　　处：《微型机与应用》2017年第24期51-53,共3页Microcomputer & Its Applications

基　　金：国家自然科学基金(61502385);国家自然科学基金(61673318);西安市科技计划项目(CXY1509(13));西安理工大学教学研究重点项目(xjy1670)

摘　　要：传统的天气状态识别往往利用许多传感器收集数据判别天气状态。然而利用图像进行天气状态识别的研究却少之又少。利用词袋模型和空间金字塔匹配对室外图像的天气状态进行识别,该方法通过分析晴天与阴天两类天气状态对图像的影响,将两类天气状态看成两种场景,对图像提取SIFT(旋转不变描述子)特征,利用词袋模型和空间金字塔匹配得到金字塔特征,然后利用金字塔特征训练分类器进而识别待测样本。在分类器构造方面,利用支持向量机(SVM)构造一级分类器,利用随机森林构造二级分类器,对测试样本经过一级分类器其介于两个支持向量之间的样本输入到二级分类器进行识别。通过对两类天气图像集的一万张图像进行测试,其识别率可以达到82%左右。Traditional weather state recognition often uses a lot of data collected by the sensors to determine the state of the weather. However, research conducted by the image recognition of the state of the weather are rare. In this paper we use bag of words model and spatial pyramid matching outdoor images to identify the state of the weather. The method which analyzes the influence of sunny and cloudy weather conditions for the two types of images, takes two types of weather conditions as the two scenarios, extracts the image SIFT （rotation invariant descriptors） feature of image, uses bag of words model and spatial characteristics of pyramid to get matching pyramid, and then uses the features of the training pyramid classifier to identify the test sample. Classifier construction aspects described herein we use Support Vector Machine （SVM） to construct a classifier, and uses random forest classifier to constructe two of the test sample through a sample input classifier interposed between two support vectors to two class classifier for recognition. Testing results show that its recognition rate can reach 82%.

关 键 词：图像分类 SIFT特征 空间金字塔 支持向量机(SVM) 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]