基于Stacking模型的红枣品种分类识别

Classification and Recognition of Jujube Varieties Based on Stacking Model Fusion

新疆红枣品种较多,在进行红枣加工过程中需要对不同品种进行区分。针对当前人工分类效率低成本高、机械分类难以确保综合品质的问题,提出了基于Stacking模型融合的红枣品种分类识别方法。试验采集5个品类的红枣图像11 280张,进行预处理,建立数据集。构建以VGG 16、ResNet 50、Densenet 121 3种不同的卷积神经网络为基学习器,逻辑回归为次级学习器的Stacking集成学习模型,进行了集成模型与单一神经网络模型以及不同基学习器组合的集成模型间的对比试验。结果表明:在红枣分类识别任务中,采用单一模型的最高准确率为88.30%,该研究提出的融合模型能够达到92.38%的准确率,分类准确率提升了4.60个百分点。

There are many varieties of jujube in Xinjiang, so it is necessary to distinguish different varieties before processing.Aiming at the problems of low efficiency and high cost of manual classification and difficult to ensure the comprehensive quality of mechanical classification, a classification and identification method of jujube varieties based on Stacking model fusion was proposed.11 280 jujube images of 5 categories were collected and preprocessed to establish a data set.A model that three different convolution neural network(VGG 16,ResNet 50,DenseNet 121) as base learner and logistic regression as secondary learner were constructed.The comparative experiments were carried out between the integrated model and single neural network model, as well as the integrated model with different combinations of base learners.The results showed that the accuracy of the proposed Stacking model was 92.38%,which was improved by 4.60 percentage points compared with that of the best single model(88.30%).This method can effectively improve the identification accuracy of jujube varieties and provide reference for Xinjiang jujube from manual sorting to automatic machine identification.