基于卷积神经网络结合极限学习机的办公垃圾图像分类

Office Garbage Image Classification Based on Convolutional Neural Networks Combined with Extreme Learning Machine

目前,基于神经网络的垃圾分类方法大多利用迁移学习获得较好的性能,然而这些方法存在训练时间长、泛化能力有限的问题。针对上述问题,本文提出一种经典网络结合极限学习机的级联网络模型应用于垃圾分类。首先,分别采用VGG19、ResNet18网络的卷积部分作为主干网络提取垃圾图片特征;其次,使用极限学习机作为分类器进行垃圾分类;最后,进行了极限学习机隐含层结点个数选择和全连接层保留与否的实验。实验结果表明,在大类的办公垃圾分类中,级联网络的训练时间和泛化能力明显优于迁移学习,同时当经典网络去除全连接层后,再结合级联网络分类可以获得更好的分类性能。

At present,most of the garbage classification methods based on neural network use transfer learning to obtain better performance.However,these methods have the problems of long training time and limited generalization ability.To solve the above problems,this paper proposes a cascade network model of classical network combined with limit learning machine,which is applied to garbage classification.Firstly,the convolution part of vgg19 and resnet18 networks are used as the backbone network to extract the characteristics of spam images;Secondly,limit learning machine is used as a classifier to classify garbage;Finally,experiments are carried out to select the number of hidden layer nodes and whether the full connection layer is retained or not.The experimental results show that the training time and generalization ability of cascaded network are significantly better than that of transfer learning.At the same time,when the classical network removes the full connection layer,combined with cascaded network classification,better classification performance can be obtained.