利用花生荚果图像特征识别品种与检验种子

为了验证以花生荚果图像特征识别品种和检验种子的可行性，选用代表北方大花生主推区的20份花生品种，从扫描图像获得花生荚果形态、颜色及纹理等50个特征，综合运用主分量分析(PCA)、神经网络(ANN)、支持向量机(SVM)、聚类分析等手段，讨论了品种识别、产地识别、DUS性状的选取和品种聚类过程，研究发现，经PCA优化特征的SVM识别模型识别效果好且识别结果稳定，20个品种的品种识别率达到90%以上。模型对3个不同产地的花生荚果正确识别率达到100%。另外从中筛选出一批对品种特异性、一致性和稳定性(DUS)测试有意义的备选特征，并建立了花生品种的谱系聚类树。研究结果对DUS性状的优选、花生品种及产地的识别及对花生谱系研究有一定参考价值。

Variety Identification and Seed Test by Peanut Pod Image Characteristics

To verify the feasibility of peanut variety recognition and seed testing by pod image characteristics, we screened 20 peanut varieties mainly released in North peanut regions and collected 50 traits based on pod morphology, colour and texture. We used PCA data optimization, neural networks, support vector machine, and clustering analysis to discuss the varieties identifica-tion, origin recognition, DUS characters selecting method and vvarieties clustering process. It has been discovered that the PCA optimization SVM model is better and its identification effect is stable. By this model, the variety recognition rate was above 90% for 20 vvarieties, and the correct origin recognition rate of three origins reached 100%. Additionally, we sorted out some useful traits for seeds DUS test from the 50 features and established the dendrogram of 20 peanut varieties. The results of this study pro-vided some references valuable to the selection of DUS traits, peanuts varieties, origin recognition, and peanut pedigree research.