Microplastic Spectral Classification Using Deep Learning with Denoising and Dimensionality Reduction

Microplastic pollution poses a significant environmental threat, especially in marine ecosystems. It is necessary to have advanced techniques for accurate identification and classification. In this paper, we present two approaches to classifying microplastics from noisy Fourier transform in-frared spectra. The proposed approaches are constructed by combinations of an Autoencoder, a dimensionality-reduction method, and a deep-learning classification model. By using a dimensionality-reduction method, the proposed approaches achieve lower computational complexity than a traditional deep-learning classification model does while offering better or comparable classification performances. Experimental results show and compare the performances of the proposed approaches on six dimensionality-reduction methods and various levels of noise.