Classification and Identification of Plastic with Laser-Induced Fluorescence Spectroscopy Based on Back Propagation Neural Network Model
WANG Xiang1, 2, ZHAO Nan-jing1*, YIN Gao-fang1, MENG De-shuo1, 3, MA Ming-jun1, 3, YU Zhi-min4, SHI Chao-yi4, QIN Zhi-song1, 5, LIU Jian-guo1
1. Key Laboratory of Optics and Technology, Anhui Institute of Optics and Fine Mechanics, Chinese Academy of Sciences, Hefei 230031, China
2. University of Science and Technology of China, Hefei 230026, China
3. Wanjiang New Industry Technology Development Center, Tongling 244000, China
4. Department of Biological and Environmental Engineering, Hefei University, Hefei 230601, China
5. Institute of Computer and Information Security, Guilin University of Electronic Technology, Guilin 541004, China
Abstract：With the advantages of low cost, good quality, strong plasticity, plastics are widely used in industrial production and daily life. However, waste plastics are prone to environmental pollution and secondary hazards without being handled properly. Recycling is expected to be a silver bullet to solve the problem of waste plastics, with the premise of accurate classification. Traditional sorting methods of waste plastics are time consuming, inefficient, and difficult to classify rapidly and effectively. Laser-induced fluorescence technique is usually used for rapid identification and quantitative analysis of organic pollutants such as oil and polycyclic aromatic hydrocarbons in water and soil with simple operation, high detection efficiency and little sample usage. It can be used to quickly collect the fluorescence spectra of different plastics, combined with the corresponding pattern recognition algorithm, the rapid and accurate identification of plastic materials can be realized. In this study, 358 sets of fluorescence spectra from eight kinds of plastics (ABS, HDPE, PA66, PLA, PP, PET, PS, PVC) were collected. A spectral matrix of 358×10 was constructed based on the characteristic peak of the spectra. and then it was processed by the method of principal component analysis, after that the linear correlation in the original spectral matrix was reduced and the accuracy of the data was improved. The results show that the cumulative variance contribution of the first three principal components was 98.085%, which was enough to characterize the main information of the original spectral matrix. Spectral classification was performed using the principal components PC1, PC2, and PC3 as inputs. Among them, the spectral polymerization degree of the same kind of plastic was high, and plastics composed with different elements such as PA66, PLA, HDPE, and PVC have better spectral resolution, while plastics containing the same elements such as PET and PLA have poor spectral resolution. The PCA algorithm is not accurate enough to identify unknown plastics. BP-Neural network was widely used in pattern recognition and classification research. The simplified feature matrix obtained by the PCA algorithm was used as the input set of the BP-neural network algorithm. Among them, 256 sets of data were randomly selected as the training set of the BP-neural network model, and the remaining 102 sets of data were used as detection sets. The value of the hidden layer of the BP neural network was set to 1, while the bipolar Sigmoid function was selected as activation function. Eight plastics were set as the output layer. The results showed that only one set of HDPE spectra in the 102 sets of spectra was misidentified as PS, and the remaining 101 sets of data were all correctly identified. The total recognition accuracy of the fluorescence spectra of eight plastics was 99%. So the laser-induced fluorescence technology combined with BP-neural network algorithm can be used to quickly and accurately identify different plastics. This study provided a new reference for automated intelligent sorting of waste plastics, reducing recycling costs and lowering the risk of waste plastics.
王 翔，赵南京，殷高方，孟德硕，马明俊，俞志敏，石朝毅，覃志松，刘建国. 基于反向传播神经网络的激光诱导荧光光谱塑料分类识别方法研究[J]. 光谱学与光谱分析, 2019, 39(10): 3136-3141.
WANG Xiang, ZHAO Nan-jing, YIN Gao-fang, MENG De-shuo, MA Ming-jun, YU Zhi-min, SHI Chao-yi, QIN Zhi-song, LIU Jian-guo. Classification and Identification of Plastic with Laser-Induced Fluorescence Spectroscopy Based on Back Propagation Neural Network Model. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2019, 39(10): 3136-3141.
[1] Tang Z W, Zhang L Z, Huang Q F, et al. Ecotoxicology and Environmental Safety，2015, 122: 343.
[2] Liao Q, Qu J H, Wang J P, et al. World Sci-Tech R&D, 2015, 37(2): 206.
[3] Rochman C M, Browne M A, Halpern B S, et al. Nature，2013, 494(7436): 169.
[4] China Plastics Processing Industry Association(中国塑料加工工业协会). China Plastics Processing Industry(2016)(中国塑料加工工业(2016)). China Plastics(中国塑料), 2017, (5): 1.
[5] Ministry of Commerce of the People’s Republic of China(中华人民共和国商务部). 2017 Report of China Renewable Resource Recycling Industry Development(中国再生资源回收行业发展报告2017). Resource Recycling(资源再生), 2017, (5): 16.
[6] YUAN Zhi-ye, BAI Gen-yun, ZHANG Wen-tao, et al(袁志业, 白根云, 张文涛, 等). Environmental Engineering（环境工程）, 2015, (S1): 615.
[7] Unnikrishnan V K, Choudhari K S, Kulkarni S D, et al. RSC Advances, 2013, 3(48): 25872.
[8] YIN Feng-fu, YAN Lei, HAN Qing-xin, et al(尹凤福, 闫 磊, 韩清新). Environmental Engineering（环境工程）2017, 35(12): 134.
[9] Wang Q, Wu X M, Chen L C, et al. Applied Spectroscopy, 2017, 71(11): 2538.
[10] HAN Xiao-shuang, LIU De-qing, LUAN Xiao-ning, et al（韩晓爽, 刘德庆, 栾晓宁, 等）. Spectroscopy and Spectral Analysis（光谱学与光谱分析）, 2016, 36(2): 445.
[11] Trost J, Zigan L, Leipertz A, et al. Applied Optics, 2013, 52(33): 8001.
[12] Taketani F, Kanaya Y, Nakamura T, et al. Journal of Aerosol Science，2013, 58: 1.
[13] YUAN Jin-sha, SHANG Hai-kun(苑津莎, 尚海昆). Electric Power Automation Equipment(电力自动化设备), 2013, 33(6): 27.
[14] DONG Zao-peng, LIU Tao, WAN Lei, et al(董早鹏, 刘 涛, 万 磊, 等). Chinese Journal of Scientific Instrument(仪器仪表学报), 2015, 36(4): 863.
[15] CHANG Liang, DENG Xiao-ming, ZHOU Ming-quan，et al(常 亮, 邓小明, 周明全，等). Acta Automatica Sinica(自动化学报), 2016, 42(9): 1300.
