国家电网故障检测多模态数据集

国家电网故障检测多模态数据集是一套专为电力领域垂直模型设计的高质量多模态数据集合。该数据集的构建旨在满足电力系统故障检测和维护的精确需求，通过集成多种数据模态，为电力系统的智能化监控和故障诊断提供全面的数据支持。数据集的来源广泛，涵盖了自然场景图像，并融合了丰富的多标签语义信息。它由五个主要的数据模态组成：视频、音频、文本、深度图和热红外图像。这些模态数据涉及变电、输电、杆塔、绝缘子等多个电力领域的关键语义概念，确保了数据集在电力系统故障检测中的应用广泛性和深度。数据采集方法多样，包括传感器、摄像头、无人机以及人工巡检等多种方式，确保了数据的全面性和多样性。具体数据量方面，图片模态拥有51237条数据记录，涵盖了1546个不同的语义概念。视频模态包含26057条数据，覆盖了1131个语义概念。文本模态包含38632条数据，涉及1604个语义概念。热红外模态的数据量为14325条，包含1267个语义概念。音频模态则有3314条数据，涉及1065个语义概念。为保证数据质量，所有数据均使用高质量格式保存，确保了数据在分析和模型训练中的准确性和可靠性。该数据集不仅为电力系统的故障检测提供了强大的数据支持，也为相关领域的研究提供了宝贵的资源，有助于推动电力系统智能化和自动化技术的发展。

The State Grid Fault Detection Multimodal Dataset is a high-quality multimodal dataset specifically designed for vertical models in the power sector. This dataset is constructed to meet the precise requirements for fault detection and maintenance of power systems, providing comprehensive data support for intelligent monitoring and fault diagnosis of power systems by integrating multiple data modalities. The dataset has extensive sources, covering natural scene images and integrating rich multi-label semantic information. It consists of five main data modalities: video, audio, text, depth maps, and thermal infrared images. These modal data involve multiple key semantic concepts in the power field such as power transformation, power transmission, transmission towers, and insulators, ensuring the wide applicability and sufficient depth of the dataset in power system fault detection. Various data collection methods are adopted, including sensors, cameras, unmanned aerial vehicles (UAVs), manual inspections and other approaches, to guarantee the comprehensiveness and diversity of the data. In terms of specific data volume, the image modality has 51237 data records covering 1546 distinct semantic concepts; the video modality contains 26057 data records covering 1131 semantic concepts; the text modality includes 38632 data records involving 1604 semantic concepts; the thermal infrared modality has 14325 data records containing 1267 semantic concepts; and the audio modality has 3314 data records involving 1065 semantic concepts. To ensure data quality, all data are saved in high-quality formats, ensuring accuracy and reliability for data analysis and model training. This dataset not only provides strong data support for power system fault detection but also offers valuable resources for research in related fields, contributing to the development of intelligent and automated technologies for power systems.

搜集汇总

背景与挑战

背景概述

国家电网故障检测多模态数据集是一套高质量的多模态数据集合，专为电力系统故障检测设计，包含视频、音频、文本、深度图和热红外图像五种模态数据，覆盖变电、输电等多个电力领域的关键语义概念。数据集总数据量为43.39GB，由国电南瑞科技股份有限公司创建，适用于电力系统智能化和自动化技术的研究。