基于多特征融合的花卉种类识别研究

发布时间：2018-05-30 16:20

本文选题：花卉图像 + 种类识别　；参考：《北京林业大学学报》2017年04期

【摘要】：花卉种类识别作为植物自动分类识别的重要分支,有着很高的研究和应用价值。针对当前花卉特征描述存在的局限和花卉识别准确率较低的实际情况,以花卉图像为研究对象,首先对复杂背景图像采用基于显著性检测的Grab Cut分割算法进行预处理,得到单一背景图像;然后在提取花卉图像花冠(所有花瓣)颜色和形状特征的基础上,创新性地提取花蕊区域的颜色和形状所包含的特征信息,并将提取到的18个特征融合成单一特征向量。以支持向量机(SVM)算法为基础构建分类器,通过实验确定核函数与最佳参数;对360幅自建花卉样本库(24个种类,每个种类15幅)进行训练和测试,其中240幅作为训练样本,120幅作为测试样本,并与基于不同特征组合的识别方法进行比较。结果表明:本文提出的基于多特征融合的识别方法具有较高的识别准确率,识别率可以达到92.50%。对通用花卉样本库Oxford 17 flower进行训练与测试,选取其中340幅作为训练样本,170幅作为测试样本,取得了较好的识别效果,验证了本文方法的有效性。
[Abstract]:As an important branch of plant automatic classification recognition, flower species recognition has high research and application value. In view of the limitation of current flower feature description and the low accuracy of flower recognition, the Grab Cut segmentation algorithm based on saliency detection is used to preprocess the complex background image. A single background image is obtained; based on the color and shape features of the Corolla (all petals) of the flower image, the feature information contained in the color and shape of the stamen region is innovatively extracted. The extracted 18 features are fused into a single feature vector. Based on the support vector machine (SVM) algorithm, a classifier is constructed, and the kernel function and optimal parameters are determined by experiments, and 360 self-built flower samples (24 species, 15 samples per species) are trained and tested. 240 samples were used as training samples and 120 samples were used as test samples, which were compared with the recognition methods based on different feature combinations. The results show that the method based on multi-feature fusion has high recognition accuracy and the recognition rate can reach 92.50%. The general flower sample library Oxford 17 flower was trained and tested, of which 340 samples were selected as test samples. Good recognition results were obtained, and the effectiveness of this method was verified.
【作者单位】：北京林业大学信息学院;北京林业大学工学院;
【基金】：国家自然科学基金项目(11272061)
【分类号】：TP391.41

【相似文献】