Fault-Level Grading of Photovoltaic Cells Employing Lightweight Deep Learning Models

Table 10

The customized CNN classification results: (a) binary classification, (b) binary classification with 0.5 as threshold, and (c) multiclass classification.

		Actual class
		Healthy		Full faulty
(a) Binary classification
Predicted class	Healthy	249		38
	Full faulty	22		85
Classwise accuracy		91.9%		69.1%
Overall accuracy		84.8%
(b) Binary classification with 0.5 as threshold
Predicted class	Healthy	247		38
	Full faulty	24		85
Classwise accuracy		91.1%		69.1%
Overall accuracy		84.3%
(c) Multi-class classification
		Healthy	Less faulty	Medium faulty	Full faulty
Predicted class	Healthy	194	4	1	4
	Less faulty	19	38	1	7
	Medium faulty	6	1	12	11
	Full faulty	7	2	2	85
Classwise accuracy		85.8%	84.4%	75%	79.4%
Overall accuracy		83.5%