万广, 陈忠辉, 方洪波, 等. 基于特征融合的随机森林模型茶鲜叶分类[J]. 华南农业大学学报, 2021, 42(4): 125-132. DOI: 10.7671/j.issn.1001-411X.202012006
    引用本文: 万广, 陈忠辉, 方洪波, 等. 基于特征融合的随机森林模型茶鲜叶分类[J]. 华南农业大学学报, 2021, 42(4): 125-132. DOI: 10.7671/j.issn.1001-411X.202012006
    WAN Guang, CHEN Zhonghui, FANG Hongbo, et al. Classification of fresh tea leaf based on random forest model by feature fusion[J]. Journal of South China Agricultural University, 2021, 42(4): 125-132. DOI: 10.7671/j.issn.1001-411X.202012006
    Citation: WAN Guang, CHEN Zhonghui, FANG Hongbo, et al. Classification of fresh tea leaf based on random forest model by feature fusion[J]. Journal of South China Agricultural University, 2021, 42(4): 125-132. DOI: 10.7671/j.issn.1001-411X.202012006

    基于特征融合的随机森林模型茶鲜叶分类

    Classification of fresh tea leaf based on random forest model by feature fusion

    • 摘要:
      目的  解决机采茶鲜叶中混有不同等级的茶叶,且混杂度高、物理特征分类精确度低的问题。
      方法  利用随机森林分类模型,提出一种基于颜色和边缘特征融合的方法。试验采集3种不同等级的茶鲜叶,对原始图像进行裁剪、尺寸归一化和去噪等处理,再进行颜色特征和边缘特征提取。通过参数的修改和测试,构建最优的随机森林分类模型,并且同K最近邻、SVM分类器进行对比试验。
      结果  特征融合之后随机森林模型的分类准确率达到99.45%,比单一颜色特征和边缘特征的分类准确率分别高7.14和9.34个百分点;比K最近邻和SVM分类器准确率分别高15.38和5.49个百分点。
      结论  所建立的方法能够对茶鲜叶单芽、一芽一叶、一芽二叶进行精确的分类。

       

      Abstract:
      Objective  To solve the problems of the machine-picked fresh tea leaves mixing with different grades of tea leaves, high mixing degree and low classification accuracy of physical characteristics.
      Method  Using the random forest classification model, a method based on the fusion of color and edge feature was proposed. We collected three different grades of fresh tea leaves, and processed the original images with cropping, size normalization and denoising, and then extracted the color features and edge features. Through parameter modification and testing, the optimal random forest classification model was constructed, and the comparison experiment was performed with the K-nearest neighbor and SVM classifier.
      Result  After feature fusion, the classification accuracy of random forest model reached 99.45%, which was 7.14 and 9.34 percentage points higher than those of single color feature and single edge feature, 15.38 and 5.49 percentage points higher than those of K-nearest neighbor model and SVM classifier respectively.
      Conclusion  The established method can accurately separate single bud, one bud and one leaf, and one bud and two leaves of fresh tea leaves.

       

    /

    返回文章
    返回