Power-Synth: Synthetic Dataset for Power Line Inspection

Accurately detecting power line defects under diverse weather conditions is crucial for ensuring power grid reliability and safety. Existing power line inspection datasets, while valuable, often lack the diversity needed for training robust machine learning models, particularly for adverse weather scenarios like fog, rain, and nighttime conditions. This paper addresses this limitation by introducing a novel framework for generating synthetic power line images under diverse weather conditions, thereby enhancing the diversity and robustness of power line inspection systems. The proposed approach employs a combination of novel heuristic image processing techniques, and a multi-domain Generative Adversarial Network (GAN) called StarGAN-v2. Initial transformations using heuristic methods simulate rain, fog, and night conditions, providing a foundation for the GAN to learn accurate mappings between weather domains. The StarGAN-v2 model, achieving its best performance with a latent dimension of 16, yielded a Frechet Inception Distance (FID) score of 24.72 and a Learned Perceptual Image Patch Similarity (LPIPS) score of 0.37 for fog, indicating high fidelity and perceptual similarity to real images. Furthermore, the impact of incorporating these synthetic images into the training process of various object detection models is thoroughly examined. The results show that models trained on a combination of synthetic and real data outperform those trained solely on either real data only or synthetic data only.

搜集汇总

背景与挑战

背景概述

Power-Synth是一个专为电力线检测设计的合成数据集，旨在通过生成雾、雨和夜间等多样化天气条件下的图像，弥补现有数据集的不足，以提升机器学习模型的鲁棒性。该数据集包含合成图像及其YOLOv8格式的边界框标签，适用于电力与能源、智能电网和图像处理等领域的研究。