Identification and visualisation of microplastics via PCA to decode Raman spectrum matrix towards imaging

To visualise microplastics and nanoplastics via Raman imaging, we need to scan the sample surface over a pixel array to collect Raman spectra as a matrix. The challenge is how to decode this spectrum matrix to map accurate and meaningful Raman images. This study compares two decoding approaches. The first approach is used when the sample contains several known types of microplastics whose standard spectra are available. We can map the Raman intensity at selected characteristic peaks as images. In order to increase the image certainty, we employ a logic-based algorithm to merge several images that are simultaneously mapped at several characteristic peaks to one image. However, the rest of the signals other than the selected peaks are ignored, meaning a low signal-noise ratio. The second approach for decoding is used when samples are complicated and standard spectra are not available. We employ principal component analysis (PCA) to decode the spectrum matrix. By selecting principal components (PC) and generating PC score curves to mimic the Raman spectrum, we can justify and assign the suspected items to microplastics and other materials. By mapping the PC loadings as images, microplastics and other materials can be simultaneously visualised. We analyse a sample containing two known microplastics to validate the effectiveness of the PCA-based algorithm. We then apply this method to analyse "unknown" microplastics printed on paper to extract Raman spectra from the complicated background and individually assign the images to paper fabric/additive, black carbon and microplastics, etc. Overall, the PCA-based algorithm shows some advantages and suggests a further step to decode Raman spectrum matrices towards machine learning.