Hybrid perovskites thin films morphology identification by adapting multiscale-SinGAN architecture, heat transfer search optimized feature selection and machine learning algorithms

Abstract

The automation in image analysis while dealing with enormous images generated is imperative to deliver defect-free surfaces in the optoelectronic area. Five distinct morphological images of hybrid perovskites are investigated in this study to analyse and predict the surface properties using machine learning algorithms. Here, we propose a new framework called Multi-Scale-SinGAN to generate multiple morphological images from a single-image. Ten different quality parameters are identified and extracted from each image to select the best features. The heat transfer search is adopted to select the optimized features and compare them with the results obtained using the cuckoo search algorithm. A comparison study with four machine learning algorithms has been evaluated and the results confirms that the features selected through heat transfer search algorithm are effective in identifying thin film morphological images with machine learning models. In particular, ANN-HTS outperforms other combinations : Tree-HTS, KNN-HTS and SVM-HTS, in terms of accuracy,precision, recall and F1-score.