For fast and accurate non-destructive differentiation of plastic pellets with similar appearance during import and export inspections, as well as for quality control purposes. Terahertz technology offers a non-destructive analysis approach for samples from diverse fields. By integrating deep learning algorithms with terahertz time-domain spectroscopy (THz-TDS) and linear discriminant analysis (LDA), it is possible to establish a highly accurate terahertz spectroscopy qualitative identification model. Additionally, the incorporation of principal component analysis (PCA) enables the application of this model to higher dimensional data. Notable differences in absorption coefficient, phase difference, and refractive index are observed among different plastic particles. The implementation of this rapid, accurate, and non-destructive detection method can greatly facilitate the testing and identification of plastic particles in customs and quality inspection departments.