Enhancing Photovoltaic Module Fault Diagnosis with Unmanned Aerial Vehicles and Deep Learning-Based Image Analysis

Table 6

Specification of pretrained network considered.

Pretrained network	Number of layers	Number of learnable parameters	Size of network	Input image size	Key features
VGG19	19	138 million	549 MB		Deep architecture with small 3×3 convolutional filters
VGG16	16	138 million	528 MB		Deep architecture with small 3×3 convolutional filters
AlexNet	8	60 million	233 MB		Use of ReLU activation functions and dropout regularization
ResNet50	50	25.6 million	102 MB		Introduction of residual connections to solve the vanishing gradient problem
DenseNet 201	201	20 million	80 MB		Densely connected convolutional neural network with efficient memory usage
GoogleNet	22	6.8 million	27 MB		Use of inception modules for efficient use of computational resources