基于卷积神经网络的牡丹花品种识别

Convolutional Neural Networks for Peony Flower Species Recognition

牡丹花品种识别对牡丹花栽培、鉴赏和科普教育具有重要意义。基于深度学习在图像分类领域的优异性能,研究基于图像分类技术的牡丹花品种自动识别方法。首先在自然条件下采集11个品种的11624幅牡丹花图像自建数据集,然后在ResNet模型框架下,修改最顶端的全连接层与分类输出层组成卷积网络主体结构,并采用数据增强和Dropout技术防止过拟合。实验结果验证了卷积神经网络在牡丹花品种图像识别任务上的优越性能,测试集上的卷积神经网络识别准确率达到98%。

The identification of peony flower species is of great significance to the cultivation,appreciation and science education of peony flower.Based on the excellent performance of deep learning technique in the field of image classification,this paper studies the automatic recognition method of peony flower species based on image classification technology.Firstly,11624 peony flower images dataset were collected under natural conditions,which consists of 11 peony flower species.Then,under the framework of ResNet model,five convolutional and pooling layers were combined as the main structure of convolutional network,and Data augmentation and Dropout techniques were adopted to prevent data overfitting.The experimental results verify the superior performance of the convolutional neural networks on the image recognition task of peony flower species.The recognition accuracy of proposed simple convolutional neural network model on the test set reaches 80%.