玉米品种图像识别中的影响因素研究

为了研究玉米品种图像识别中的关键影响因素,搭建了一套基于PCA和ICA特征提取和支持向量机(SVM)分类算法的玉米品种识别系统,采用扫描仪获得了11个品种每个品种50粒图像,基于图像的像素特征和统计特征,分别研究了主分量分析(PCA)和独立分量分析(ICA)的特征提取和特征优化方法,并进一步考察了支持向量机(SVM)模式分类过程中的关键参数优化问题.试验结果表明,对11个品种550个籽粒的品种最高检出率为97.17％,在同样的情况下ICA优化的特征较PCA优化的特征识别率能提高3％左右,适当选择统计特征比使用像素特征识别率提高约10％,另外SVM参数影响到识别效果,但整体影响不大.本方法与结论对玉米种子纯度和品种真实性检验具有积极意义.

Study on the Influencing Factors of Maize Cultivars by Image Classification

In order to research the key influencing factors in the corn varieties image recognition,we build a corn varieties recognition system based on PCA,ICA feature extraction and support vector machine(SVM)classification algorithm.11 varieties,each variety 50 image were taken with scanners.Based on pixel features and statistical characteristics of these images,some feature extraction and features optimization methods by PCA and ICA were studied respectively.And also,we inspected the key parameter optimization process in pattern classification based on support vector machine(SVM).Test results showed that the highest rate of varieties is 97.17% for 11 varieties of 550 kernels.In the same way,the recognition rate of kernels optimized by ICA characters is about 3% higher than PCA-based optimization.Recognition using statistical characteristic appropriate selection is about 10% higher than pixels features.In addition,the SVM parameters influenced the identifying effect,but the overall effect is not big.The method and the conclusion of this paper have positive significance for corn seed purity and varieties test.