Qualitative analysis for microplastics based on GAF coding and IFCNN image fusion enabled FITR spectroscopy method

Qualitative analysis for microplastics is very important since it represents a looming threat to our environment and healthy. FTIR is a powerful tool for microplastics classification, however, it is challenging to build excellent performance model because of their spectra are highly complex. This study proposed a GAF coding combing with IFCNN image fusion methodology for the first time to address the FTIR qualitative analysis of microplastics. Based on the international available microplastic FITR dataset, GAF coding was first performed to transform original FTIR spectra into GASF and GADF images. Then IFCNN network was designed to fuse these two types of GAF images, and finally CNN network was implemented to establish the qualitative model of the fused images to successfully classify fourteen types of microplastics. The experimental results show that IFCNN fusion method has transcended other three image fusion methods of Wavelet fusion, IHS fusion and Bayesian fusion with the best image evaluation indicators of AG, EN, SF, NMI and PSNR. The results of GASF-CNN and GADF-CNN classification models prove that the model performance has been greatly improved by converting 1D spectra into 2D GAF images. Furthermore, by fusing the GASF and GADF images, the IFCNN-CNN has achieved the most excellent qualitative model with the highest values of accuracy and sensitivity of 0.992 and 0.859, respectively, which illustrate that the fused image outperformed single GASF and GADF image. The proposed GAF-IFCNN-CNN method has significantly surpassed traditional classification models of 1DCNN, KNN, SVC and SVC-MCCV, which has increased the classification accuracy by 13.24%, 9.61%, 8.65%, and 4.53%, respectively and has increased the sensitivity by 18.94%, 18.03%, 4.65%, and 9.10%, respectively. This work demonstrates the potential of the GAF coding and IFCNN image fusion enabled FTIR spectroscopy technology to be used as a rapid, accurate and reliable detection method for a variety of microplastics.