摘要
花卉识别在自动化种植,机械采摘,病虫害防治,鲜花定级等方面均起到关键性作用。为了进一步提高卷积神经网络在花卉识别领域的准确率,尤其是提高花卉被局部遮挡情况下的识别准确率。提出了一种基于花蕊区域定位的花卉识别方法,通过目标检测算法Faster-RCNN对图像中的花蕊区域进行定位,再通过花蕊的特征进行种类识别。通过对牛津大学Flowers102数据集中的12种花卉进行验证,基于花蕊区域定位的识别准确率可以达到96.07%,高于小型网络Lenet-5,与大型网络Vgg-16及Inception-V3识别准确率相近,验证了花蕊区域可以提供足够的特征进行识别。对于花瓣高度遮挡的情况,提取整幅图像特征的传统卷积神经网络Vgg-16和Inception-V3的识别准确率大幅下降至25.33%和35.14%,而基于花蕊区域定位的识别准确率可以达到88.93%。表明该方法有效的提升了花卉被局部遮挡情况下的识别准确率,提高了抗遮挡能力。
Flower recognition plays a key role in automatic planting, mechanical picking, pest control, and flower grading. In order to improve the accuracy of the convolutional neural network in the field of flower recognition. Especially improve the recognition accuracy of flowers under partial occlusion. In this paper, a flower recognition method based on the localization of the stamen area is proposed. The target detection algorithm Faster-RCNN is used to locate the stamen area in the image, and then the species identification is carried out through the characteristics of the stamen area. Verification of 12 species of flowers in the Oxford Flowers102 dataset show that the accuracy of the positioning of the stamen area can reach 96.07%,higher than small network lenet-5, similar to large network vgg-16 and Inception-V3. It is verified that the stamen area can provide sufficient features for identification. For the case of high occlusion of the petals. The recognition accuracy of traditional convolutional neural networks Vgg-16 and Inception-V3, which extract the whole image feature, drop sharply to 25.33% and 35.14. the recognition accuracy base on stamen positioning could reach 88.93%. Show that this method can effectively improve the recognition accuracy of flowers under partial occlusion and improve the anti-occlusion ability.
作者
任意平
夏国强
李俊丽
Ren Yiping;Xia Guoqiang;Li Junli(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China)
出处
《电子测量技术》
2020年第7期97-102,共6页
Electronic Measurement Technology
基金
国家自然科学基金(61163051)
云南省教育厅科学研究基金(2015Y071)
关键词
花卉识别
卷积神经网络
目标检测
花蕊区域
flower recognition
convolutional neural network
target detection
stamen area
