查询结果:   李帷韬,陶海,吴四茜,王建平,徐晓冰.基于深度学习的青梅品级半监督智能认知方法[J].计算机应用与软件,2018,35(11):245 - 252.
中文标题
基于深度学习的青梅品级半监督智能认知方法
发表栏目
图像处理与应用
摘要点击数
635
英文标题
SEMI-SUPERVISED INTELLIGENT COGNITION METHOD OF GREENGAGE GRADE BASED ON DEEP LEARNING
作 者
李帷韬 陶海 吴四茜 王建平 徐晓冰 Li Weitao Tao Hai Wu Sixi Wang Jianping Xu Xiaobing
作者单位
合肥工业大学电气及自动化工程学院 安徽 合肥 230009 东北大学 辽宁 沈阳110004    
英文单位
School of Electrical Engineering and Automation, Hefei University of Technology,Hefei 230009,Anhui,China Northeastern University,Shenyang 110004,Liaoning,China    
关键词
青梅品级 半监督学习 自适应架构卷积神经网络 万局逼近能力 潜在语义熵测度
Keywords
Greengage grade Semi-supervised learning Adaptive structure convolutional neural network Universal approximation Latent semantic entropy measurement
基金项目
流程工业综合自动化国家重点实验室开放课题(PAL-N201605)
作者资料
李帷韬,副教授,主研领域:模式识别,图像处理。陶海,硕士生。吴四茜,硕士生。王建平,教授。徐晓冰,副教授。 。
文章摘要
针对水果品级监督学习认知方法中样本获取困难、特征空间充分性表征和分类器能力不足的缺陷,模仿人反复推敲比对的信息交互模式,提出一种基于深度学习的青梅品级半监督智能认知方法。基于半监督学习扩充训练样本库,从信息论角度建立多层面特征充分表征的青梅品级认知智能决策信息系统;基于具有充分表征性的自适应架构卷积神经网络和随机配置网络,建立青梅图像由全局到局部多层面充分表征特征空间和具有万局逼近能力的分类准则;基于广义熵理论,建立青梅图像认知结果的语义熵测度评价指标;构建基于不确定认知结果熵测度指标约束的特征空间和样本信度自寻优反馈调节机制。仿真实验表明,该方法青梅品级平均识别率为98.2%。
Abstract
Aiming at the shortcomings of sample acquisition, feature space adequacy and classifier capability in fruit grade supervised learning cognition methods, a semi-supervised intelligent cognition method for greengage grade based on deep learning was proposed, which imitated the information interaction mode of human repeated deliberation and comparison. Based on semi-supervised learning, a multi-level intelligence system was developed to fully characterize the intelligent decision information system of greengage grade from the perspective of information theory. Based on the self-adaptive architecture convolution neural network and random configuration network with abundant representation, the classification criterion of greengage images from global to local multi-layered was established, which fully characterized the feature space and had universal approximation. Based on generalized entropy theory, semantic entropy measurement evaluation index of greengage images recognition results was established. A self-optimizing feedback regulation mechanism of feature space and sample reliability was constructed based on entropy measurement index constraints of uncertain cognitive results. The simulation results show that the average recognition rate of greengage grade is 98.2% 
下载PDF全文