基于PCA和神经网络的荞麦剥壳混合物识别  

Recognition of Hulled Buckwheat Mixture Based on PCA and BP Neural Network

作　　者：吕少中 杜文亮[1] 陈伟[1] 刘广硕[1] 张丽杰[2] 

机构地区：[1]内蒙古农业大学机电工程学院,呼和浩特010018 [2]内蒙古工业大学信息工程学院,呼和浩特010080

出　　处：《农机化研究》2018年第1期166-170,177,共6页Journal of Agricultural Mechanization Research

基　　金：国家自然科学基金项目(31260409);内蒙古自然科学基金项目(2014MS0310)

摘　　要：针对荞麦剥壳机调节运行参数时需要对出料口混合物中各种成分进行定量分析,而传统人工分析方法耗时且主观性强的问题,研究了一种基于主成分分析和BP神经网络的荞麦剥壳混合物识别方法。采集未剥壳荞麦、已剥壳完整荞麦米和破损荞麦米的图像,对图像进行预处理后,提取了每个单独籽粒图像的12个颜色特征、10个形状特征和18个纹理特征。使用主成分分析法将40个特征参数映射为5个综合特征作为输入参数,构造了一个5-11-3结构的单隐层BP神经网络对荞麦剥壳混合物进行识别,试验结果表明:该BP神经网络对未剥壳荞麦、已剥壳完整荞麦米和破损荞麦米的识别正确率分别为98%、90%和98%,平均正确率为95%,能够对荞麦剥壳混合物进行有效的识别。Aiming at the problem that the quantitative analysis of buckwheat mixture in buckwheat huller＇s outlet when adjusting operation parameter, traditional manual analysis method is subjective and consume more time. A novel recogni- tion method based on PCA and BP neural network was studied. The images of unhulled buckwheat , complete buckwheat kernel and broken buckwheat kernel were captured, subsequently were preprocessed. 12 color feateurs, 12 shape features and 18 texture features on each grain were extracted, the 40 feature parameters were mapped to 5 principal components by principal component analysis. A BP neural network with 5-11-3 single hidden layer structure that used 5 principal components as inputs was constructed, which was used to recognize the buckwheat mixture. The results showed that the recognition rate of unhulled buckwheat , complete buckwheat kernel and broken buckwheat was 98% , 90% and 98% , respectively, and the average recognition rate was 98%. It is concluded that this method is a effective way to recognize buckwheat mixture.

关 键 词：图像处理 荞麦 识别 主成分分析 神经网络 

分 类 号：S126[农业科学—农业基础科学] TP391.41[自动化与计算机技术—计算机应用技术]