摘要
随着我国花卉产业规模扩大,养花赏花人数不断增加,如何能够快速自动地识别花卉种类及花卉病症受到了广泛的关注.目前图像识别技术已有大量研究工作,卷积神经网络研究已获重大突破,其在花卉分类问题上的应用也有很大进展,但关于花卉疾病识别方面的研究仍然较少.针对10种常见观赏花卉,提供一个包含4种花卉的共16种叶部病症的图像数据集,结合网络多输入和迁移学习方法,设计并实现基于卷积神经网络的分类模型,集成为花卉分类-病症识别一体化工具.实验结果表明,所设计的基于卷积神经网络的分类模型有较高的准确率,病症识别总体准确率达到88. 2%,经迁移学习后提升至94. 4%,相比于基于支持向量机的分类模型准确率高出至少27. 0%.
With the development of flower market and industry,the number of people who cultivate and admire flowers continues to increase.How to identify flower categories and their diseases quickly and automatically has received widespread attention,since it is extremely convenient for the flower management in home gardening or large plantations. At present,there has been a lot of research work on computer vision and image recognition. The convolutional neural network research has achieved a major breakthrough,whose application in flower classification has also made great progress. However,research on flower disease identification is still relatively insufficient,so is the data set. Aiming at 10 species of common ornamental flowers,we provide an image data set including 16 leaf diseases of four species of flowers. What’s more,we design and implement a classification model based on the convolutional neural network combining with methods such as image pre-processing,network multi-input,data augmentation,batch normalization and so on,and integrate it into a tool for flower category and disease classification. The result of the experiments shows that the designed classification model has a high illness recognition accuracy of 88. 2%,and achieved a better one of 94. 4% combining with transfer learning,which is higher at least 27. 0% than the classification model based on the support vector machine.
作者
林君宇
李奕萱
郑聪尉
罗雯波
许蕾
LIN Jun-yu;LI Yi-xuan;ZHENG Cong-wei;LUO Wen-bo;XU Lei(Department of Computer Science and Technology,Nanjing University,Nanjing 210023,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2019年第6期1330-1335,共6页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(61272080,61403187)资助
江苏省自然科学基金项目(BK20140611)资助
2018年南京大学大学生创新训练计划项目(G201810284057)资助
关键词
卷积神经网络
花卉病症图像分类
多输入
支持向量机
迁移学习
convolutional neural network
flower disease classification
multi-input
support vector machine
transfer learning
