A CNN-Based Classifier for Quality Image Selection from Sky Cameras

This paper presents quality enhancement techniques for all sky images in solar parks that uses sky camera-based systems. This study addresses the real-world challenges associated with deploying these systems, particularly focusing on external factors such as weather conditions, and interference due to soiling and dust accumulated on top of the domes of the sky cameras installed in desert conditions as well as human, birds, or insects that can compromise the quality of sky images used for irradiance estimation. To tackle these issues, a Convolutional Neural Network (CNN)-based classifier capable of identifying and excluding invalid images from the data processing pipeline is proposed. The proposed classifier achieves an accuracy exceeding 98% in detecting invalid images and is more than seven times smaller than traditionally used pre-trained models like VGG-16 and ResNet50, while also being over 16% faster. This research paves the way for more reliable operations in solar parks, leveraging ML-based methodologies for enhanced solar energy management.