基于融合卷积神经网络的花卉识别方法
Flower Recognition Based on Fusion Convolution Neural Network
段毛毛 1翟睿1
作者信息
- 1. 中国石油大学(北京)克拉玛依校区 克拉玛依 834000
- 折叠
摘要
利用计算机技术能够帮助人们快速识别不同种类的植物.为解决复杂背景下植物识别困难的问题,选取简单和复杂两种不同背景条件的花卉数据集为研究对象,从花卉图像的有效特征与无效特征出发,首先以花卉图像的全局特征为输入,采用多种卷积神经网络对植物进行分类识别,寻找最佳网络模型.然后使用Mask R-CNN提取花卉图像的有效区域,去除图像中的无效区域,使模型能够获取更为精确的有效特征.最后将处理后的图像作为最佳网络模型的输入,再次对模型进行训练.实验结果表明,此方法能够使简单背景下的花卉识别的准确率提高3%,复杂背景下花卉识别的准确率提高5%.
Abstract
Computer technologies can help people identify different kinds of plants quickly.In order to solve the difficult prob-lem of plants recognition under complex background,two kinds of flower data sets are selected as the research object with different background.Firstly,a variety of convolution neural networks are used to classify flowers with global features of flower images and find the best network.Secondly,this paper extracts effective regions of the plant image and removes the invalid regions in the image by using Mask R-CNN,which makes sure that the network following can get more accurate effective features.Finally,the best net-work is trained again with processed images.The results show that this method can improve the accuracy of plant recognition with simple background by 3%and in complex background by 5%.
关键词
卷积神经网络/Mask/R-CNN/花卉识别Key words
convolutional neural network/Mask R-CNN/flower recognition引用本文复制引用
基金项目
克拉玛依市优秀科技创新人才基金项目(XQZX20230110)
出版年
2024
国家科技期刊平台
NSTL
万方数据