Aerial Imaging-Based Soiling Detection System for Solar Photovoltaic Panel Cleanliness Inspection

Area of Science:

• Renewable Energy
• Artificial Intelligence
• Computer Vision

Background:

• Solar photovoltaic (PV) panel inspection relies on Unmanned Aerial Vehicles (UAVs) with visual cameras for monitoring.
• Soiling, especially bird droppings, significantly reduces solar panel power generation and can cause hotspots.
• Detecting small, indistinct soiling like bird droppings from aerial images presents a significant challenge.

Purpose of the Study:

• To develop an AI-assisted soiling detection methodology for solar PV panels using UAV-captured RGB images.
• To create an autonomous, end-to-end soiling detection model capable of identifying common soiling types like dust and bird droppings.
• To address the specific challenges of detecting small and indistinct bird droppings in aerial imagery.

Main Methods:

• A custom AI model, SDS-YOLO (Soiling Detection System YOLO), was developed, incorporating a Convolutional Block Attention Module (CBAM).
• The model features two dedicated detection heads optimized for distinguishing between dust and bird droppings.
• A dataset of aerial RGB images featuring PV panels with dust and bird droppings was collected for training and validation.

Main Results:

• SDS-YOLO demonstrated significantly improved detection accuracy for bird droppings compared to existing YOLO models (v5, v8, v11).
• Integration of CBAM led to a 40.2% increase in mAP50 and a 26.6% F1 score improvement for bird dropping detection.
• The model also showed robust performance for dust detection, with improved feature extraction and reduced false positives due to CBAM.
• SDS-YOLO achieved a 24% reduction in parameter count, enhancing its suitability for edge computing.

Conclusions:

• The proposed SDS-YOLO model effectively detects various soiling types on solar panels, particularly excelling at identifying challenging bird droppings.
• The CBAM integration is crucial for enhancing feature extraction and improving detection accuracy, especially for small and indistinct objects.
• SDS-YOLO offers an efficient and accurate solution for solar panel inspection and monitoring, suitable for deployment on edge devices.