Effect of GAN in Unsupervised fault classification of IR Images of Photo voltaic cell using UMAP

Resumen

Classification of faults occurring in photo voltaic cell is mainly classified in to two ways
,supervised classification and unsupervised classification. The advantage of unsupervised
classification is that the target variable is not required. The IR dataset used is highly
imbalanced , to achieve balance generative adversarial network is used. In this paper
unsupervised classification is done using HDBSCAN and the input to it is obtained through
Uniform manifold approximation projection (UMAP) dimensionality reduction technique. In
earlier works the dimensionality reduction is carried out using principal component analysis,
but the number of components required in this case is more than that of the number of
components obtained through UMAP. Hence the computational cost of machine learning
classifiers can be reduced using UMAP dimensionality reduction technique in supervised
classification. In unsupervised case the computational cost incurred by hierarchical densitybased
spatial clustering of applications with noise is reduced. The performance of the
proposed method is compared with the existing methods and it is found that the proposed
method yields fruitful results.