MDFN: Enhancing Power Grid Image Quality Assessment via Multi-Dimension Distortion Feature

Area of Science:

• Computer Vision
• Artificial Intelligence
• Electrical Engineering

Background:

• Low-quality power grid images degrade deep learning performance in the power industry.
• Current blind image quality assessment (BIQA) methods often rely on single features, neglecting critical factors like noise and brightness.
• Power grid images frequently suffer from noise, underexposure, and overexposure, necessitating specialized quality assessment.

Purpose of the Study:

• To develop an advanced image quality assessment technique for power grid images.
• To address the limitations of existing BIQA methods by incorporating multi-dimensional features.
• To enhance the accuracy of screening high-quality power grid images for deep learning applications.

Main Methods:

• Proposed a multi-dimension distortion feature network (MDFN) integrating Convolutional Neural Networks (CNN) and Transformer architectures.
• Employed a dual-branch feature extractor: CNN for local features, Transformer for local and global features.
• Introduced a frequency selection module (FSM) to separate and fuse high-frequency (edges, details) and low-frequency (semantic, structural) components.
• Developed a novel method to extract and integrate noise and brightness features with the CLS token for quality prediction.

Main Results:

• The proposed MDFN method demonstrated superior performance compared to existing approaches.
• Experimental results were validated across three public datasets and a dedicated power grid image dataset.
• The network effectively captures both local and global image characteristics, along with noise and brightness distortions.

Conclusions:

• The MDFN provides a more accurate and robust solution for blind image quality assessment in the power industry.
• Integrating multi-dimensional features, including frequency components, noise, and brightness, significantly improves assessment accuracy.
• The method is effective in screening high-quality power grid images, essential for reliable deep learning model training and performance.