摘要
为实现复杂背景下准确、快速地识别多种水果,提出了基于改进DarkNet-53卷积神经网络的水果分类识别模型.该模型在DarkNet-53网络模型基础上,用组归一化方法替换原有的批量归一化方法,改进模型结构、优化参数.在此基础上,引入YOLOv3算法对图像全局信息进行目标预测,构建水果目标检测模型.从建立的水果图像库中随机抽取样本作为训练集和测试集,测试该方法性能.结果表明:所构建模型能够有效提取水果图像的不同层特征,与原模型相比不依赖于批量大小,准确率达到95.6%;使用改进的DarkNet-53作为主干网络的水果目标检测模型,平均识别精度达到85.91%.
In order to achieve accurate and fast recognition of multiple fruits under complex backgrounds,a fruit classification and recognition model based on an improved DarkNet-53 convolutional neural network is proposed.This model is based on the DarkNet-53 network model and replaces the original batch normalization method with a group normalization method to improve the model structure and optimize parameters.Based on this,the YOLOv3 algorithm is introduced to predict the global information of the image,and a fruit target detection model is constructed.Randomly draw samples from the established fruit image database as training and testing sets to test the performance of the method.The experimental results show that the fruit classification model constructed can effectively extract different layer features of the fruit image.Compared with the original model,it does not depend on the batch size,and the accuracy rate is 95.6%.The improved DarkNet-53 is used as the fruit target detection model for the backbone network.The recognition accuracy reached 85.91%.
作者
王辉
张帆
刘晓凤
李潜
WANG Hui;ZHANG Fan;LIU Xiao-feng;LI Qian(School of Information Engineering,Minzu University of China,Beijing 100081,China;Qingdao Technical College,Qingdao 266555,China)
出处
《东北师大学报（自然科学版）》
CAS
北大核心
2020年第4期60-65,共6页
Journal of Northeast Normal University(Natural Science Edition)
基金
国家自然科学基金资助项目(61672553)
教育部社科基金资助项目(18YJAZHO87).
