期刊论文

Research on pest recognition based on cascaded AdaBoost classifier

中文

中国农机化学报

2095-5553

2019

40

8

127-131

10.13733/j.jcam.issn.2095-5553.2019.08.22

国家“十二五”科技计划支撑项目（2012BAD35B00）；

本校为第一机构

农作物虫害预防是农业生产中的重要环节。针对传统的虫害预防工作强度大、耗时长、效率低的问题,本文应用机器学习理论,在农作物害虫识别方面进行相关的研究,提出一种基于级联AdaBoost分类器的虫害识别方法。使用Haar-like特征提取害虫的特征,将提取到的特征构建弱分类器,并通过AdaBoost算法将构建得到的弱分类器集合得到强分类器,最后通过级联的方式得到一个级联AdaBoost分类器来识别害虫。试验表明,本文方法对简单背景的虫害图片能够达到95.71%的识别率,对复杂背景的虫害图片能达到86.67%的识别率,为农作物虫害的识别和预防提供有效途径。

Prevention of crop pests is an important part in agricultural production.To solve the problems of high intensity,long time-consuming and low efficiency of traditional pest prevention,machine learning theory is applied to identify crop pests in this paper and put forward a pest recognition method based on cascaded AdaBoost classifier.The method uses Haar-like feature to extract pest features and using these extracted features constructing weak classifiers.Then using these weak classifier constructing strong classifier set by Adaboost algorithm.Finally,using a cascading way to get a cascade AdaB...