A continuous projection algorithm based on the number of band constraints(SPAsa)was proposed to improve the computational speed of the model.Firstly,waste plastic samples and corresponding hyperspectral data were collected,and the target plastics were labelled one by one by the cosine similarity algorithm adaptive labelling method to construct a waste plastic sample library,which was classified using the PLS-DA method.Secondly,the number of feature variables extracted by the SPASA algorithm,which could constrain the range of the number of feature wavelengths,was set,and eight feature wavelengths were optimally extracted from the plastic near-infrared full-spectrum data of 256 wavelengths within the limited range.The full-spectrum PLS-DA model,the SPA-PLS-DA model and the SPAsa-PLS-DA model were established,and the classification accuracies of the three classification models for the 12 plastics were all greater than 97％.Finally,the SPAsa-PLS-DA model was applied to the sorting system to test the classification effect and speed,and the SPAsa-PLS-DA model was able to separate the target ABS plastics from other plastics with an accuracy of up to 100％,and the running time of the sorting system to process one frame of data was less than 2 ms.
关键词
近红外光谱/高速/塑料分选/连续投影算法/偏最小二乘判别分析
Key words
near infrared spectroscopy/high speed/plastic sorting/continuous projection algorithm/partial least squares discriminant analysis
维普
万方数据