基于改进BP网络的小麦品种识别
Classification of Wheat Varieties by Improved BP Neural Network
孟惜 1王克俭 1韩宪忠1
作者信息
- 1. 河北农业大学信息科学与技术学院,河北保定071001
- 折叠
摘要
为了提高小麦品种的识别准确率,以河北农业大学选育的6个小麦品种为研究对象,对小麦籽粒图像进行中值滤波阈值分割等方法预处理后,对形态、颜色、纹理3个方面进行特征提取.其次利用BP神经网络对单个品种的小麦进行识别,然后结合主成分分析(PCA)法降维研究一次性识别多类小麦品种,最后为避免神经网络的局限性,利用PSO算法优化网络权值参数.结果表明:BP网络对单个小麦品种具有非常好的识别效果,其中河农7069品种的识别准确率达100％;结合PCA法降维后小麦品种平均的识别准确率为91.582％;利用PSO算法优化网络后识别准确率增加至94.3％,达到了更好的识别分类效果.
Abstract
In order to improve the recognition accuracy of wheat varieties,six cultivars bred by Hebei Agricultural University were selected as the research objects,the wheat grain image was pretreated by median filter and threshold segmentation,three aspects of morphology,color and texture were extracted.Secondly using BP neural network to identify wheat varieties of a single species,then one-time identification of multiple wheat varieties was studied,and PCA method was used to reduce dimensionality.Finally,in order to avoid the limitation of the neural network,the PSO algorithm was used to optimize the network weights.Results:BP network has a very good recognition effect on a single wheat variety,identification accuracy rate of Henong 7069 reached 100％;the average recognition accuracy of wheat cultivars was 91.582％ after combining with PCA;after using PSO algorithm to optimize the network,the recognition accuracy rate reached 94.3 ％,this achieves a better indentification of classification effect.
关键词
特征提取/神经网络/PSO算法/主成分分析法/识别准确率Key words
feature extraction/neural networks/PSO algorithm/principal component analysis/recognition accuracy引用本文复制引用
基金项目
河北省高等学校科学技术研究项目(ZD2016158)
出版年
2017
国家科技期刊平台
NSTL
维普
万方数据