Research On Key Technologies Of Smart Fine Classification Of Urban Plastic Household Waste

With the continuous development of human society,the use of plastic products and plastics will be more and more.The existing plastic waste sorting equipment and technology in China are still basically in the stage of sorting mixed waste plastics,which cannot meet the urgent demand for fine sorting of waste plastic materials.Coupled with the national control of imported wastes,China is forced to achieve a fine classification of waste plastics.This paper to achieve the high quality of the plastic waste recycling,this paper proposes a based on the analysis of the independent source plastic garbage spectrum feature extraction method,the method adopts the Nicolet iS10 Fourier transform infrared spectrometer for five polymers spectra collection,analysis of spectral data processing using the independent source analysis,combining with the characteristics of polymer chemical bonds in the spectral data,the stability characteristics of the extracted independent components to obtain spectra of corresponding points figure analysis comparison,implements the effective extraction of all kinds of polymer spectral characteristics.A fine classification model of plastic waste based on Fisher vector coding and deep convolutional neural network(FV-DCNN)is studied.Based on the original data of the plastic waste spectrogram,the Origin software was used to reproduce and screen the spectrogram,and then the spectrogram was denoised.Finally,the detailed classification model of plastic waste was established by combining the deep convolutional neural network with Fisher vector coding.In the model,the denoising algorithm combined with wavelet transform and bilateral filtering and multi-mode feature fv-dcnn algorithm are used.Bilateral filtering based on wavelet transform overcomes the residual noise generated by bilateral filtering in some regions after filtering.Combining with the characteristics of time-frequency localization and multi-resolution of wavelet transform,the noise elimination is realized while some details are reserved.Multimodal feature FV-DCNN uses fusion automatic learning feature extraction and Fisher vector to encode the logarithmic likelihood parameters of the model.Batch standardization can alleviate internal covariate migration and prevent gradient dispersion in deep neural network training,which is conducive to improving training efficiency.The experimental results show that the classification accuracy of this model is more than 91%,and the accuracy of this model is proved by comparison with other models.This paper provides a model method for the fine classification of plastic waste in China,and makes it possible for the fine classification of plastic waste.