基于红外热成像技术的玉米种子活力等级分类方法研究

针对传统玉米种子活力等级分类方法耗时长、环境要求严格、对种子产生损伤等问题,利用红外热成像技术结合SVM算法,建立了快速、无损、高效的玉米种子活力等级分类方法。首先采用人工老化的方法将1 200粒玉米种子分组分别老化0 h,72 h,144 h。利用不同老化时间玉米种子具有不同的生理特性,通过红外热成像仪采集温度胁迫后自然冷却的玉米种子红外热像图,提取温度值作为特征。随后对玉米种子进行标准萌发实验,根据实验结果,将玉米种子分为高活力,中活力和低活力3个活力等级。将温度值作为特征,活力等级作为标签分别建立K最近邻(KNN)和支持向量机(SVM)模型并进行训练,以模型分类准确率和训练时间作为评价指标,确定较佳模型,最终通过网格搜索对选择的模型参数进行优化。结果表明基于红外热成像技术结合支持向量机(SVM)建立的模型,训练集准确率达到了92.4%,测试集准确率为91%,训练用时0.12s。该模型经过优化后训练集准确率达到了97.1%,测试集准确率达到了96.5%。

Research on the classification method of maize seed vigor grade based on infrared thermal imaging technology

Aiming at the problems of the traditional corn seed vitality grade classification method that takes a long time, has strict environmental requirements and damage to seeds, a fast, non-destructive and efficient corn seed vitality grade classification method is established by using infrared thermography technology combined with SVM algorithm. Firstly, 1200 corn seeds were grouped into groups to age 0h, 72h and 144h respectively by artificial aging method. Using the different physiological characteristics of maize seeds at different aging times, infrared thermal images of maize seeds that were naturally cooled after temperature stress were collected by infrared thermal imager, and temperature values were extracted as characteristics. Subsequently, the maize seeds were subjected to standard germination experiments, and according to the experimental results, the maize seeds were divided into three vitality levels: high vitality, medium vitality and low vitality. The K nearest neighbor (KNN) and support vector machine (SVM) models are established and trained with temperature value as the feature and vitality level as the label, and the model classification accuracy and training time are used as evaluation indicators to determine the best model, and finally the selected model parameters are optimized by grid search. The results show that based on the infrared thermal imaging technology combined with the model established by the support vector machine (SVM), the accuracy of the training set reaches 92.4%, the accuracy of the test set is 91%, and the training time is 0.12s. After optimization, the accuracy of the training set of the model reaches 97.1%, and the accuracy of the test machine reaches 96.5%.