融合线性和非线性残差的一次性垃圾分类模型

Disposable Garbage Classification Model Combining Linear and Nonlinear Residuals

为应对一次性塑料垃圾难检测的问题,应用卷积神经网络提出一次性塑料垃圾分类模型。该模型在预处理阶段模拟手工提取方式捕获线性和非线性残差信息,在残差特征学习阶段通过卷积神经网络融合线性和非线性残差特征。仿真结果表明:线性和非线性残差特征具备较优的分类能力,深层次网络有利于融合各类残差且学习捕获高级语义特征信息,本模型的检测分类准确率为75.84%,优于传统HOG模型约8%。

In order to solve the problem of difficult detection of disposable plastic waste,a classification model of disposable plastic waste is proposed by using convolution neural network.In the preprocessing stage,the model simulates manual extraction to capture the linear and nonlinear residual information.In the learning stage of residual features,the deep convolution neural network is used to fuse the linear and nonlinear residual information.The simulation results show that the linear and nonlinear residual features have better classification ability.The deep network was conducive to the fusion of various types of residual and learning to capture advanced semantic feature information.The accuracy of classification of this model was 75.84%,which was about 8%better than that of traditional HOG model.