摘要
深度学习已广泛地应用于植物图像识别分类中.由于蘑菇图像的识别分类难度较大,针对如何提高识别模型的准确率和泛化能力,提出了单一背景下的蘑菇图像识别方法.针对传统的卷积神经网络中存在的空间冗余问题,采用了一种降梯度卷积训练模型,有效提高了蘑菇分类图像的识别性能.通过一个包含8123个样本的蘑菇数据集测试,降梯度卷积训练模型的平均耗费时间为0.985 s,第1张图像的平均命中准确率达到了91.6%,实验结果表明:降梯度卷积训练模型在单一背景、数据量较大的情况下,识别速度和准确率均优于传统的传统卷积神经网络.
Deep learning has been widely used in plant image recognition and classification.Because it is difficult to recognize and classify mushroom images,a method of mushroom image recognition under a single background is proposed to improve the accuracy and generalization ability of the recognition model.Aiming at the problem of spatial redundancy in traditional convolution neural network,a reduced gradient convolution training model is adopted,which effectively improves the recognition performance of mushroom classification image.Through a mushroom data set test with 8123 samples,the average time consumed by the reduced gradient convolution training model is 0.985s,and the average hit accuracy of the first image reaches 91.6%.The experimental results show that the recognition speed and accuracy of the reduced gradient convolution training model are better than those of the traditional convolution neural network in the case of a single background and a large amount of data.
作者
罗奇
LUO Qi(RoachSchool of Physical Education and Information Technology,Wuhan Institute of Physical Education,Wuhan 430079,China)
出处
《中国食用菌》
北大核心
2019年第6期26-29,33,共5页
Edible Fungi of China
基金
湖北省自然科学基金计划项目(2017CFB560)
关键词
深度学习
卷积神经网络
图像识别
图像频率
特征图
deep learning
convolution neural network
image recognition
image frequency
feature map
作者简介
罗奇(1982-),男,博士后,副教授,硕士研究生导师,主要从事计算机应用方面研究。
