Multi-spectral fusion power equipment fault recognition based on prompt learning

基于提示学习的电力设备故障多谱段融合识别方法

School of Computer Science, Northwestern Polytechnical University, Xi'an 710072, China

Received: 5 March 2024

Abstract

To address the issue of weak fault recognition ability of power equipment in single-spectrum images, a multi-spectral fusion recognition method based on prompt learning is proposed. A multi-spectral imaging system is used to capture images of normal and faulty power equipment, collecting multi-spectral data including visible light, infrared, and ultraviolet. The collected dataset is annotated with text labels for training the large model. The generalization ability of the large model in power equipment fault recognition is verified, and the original large model is tested on the collected dataset for device type and fault recognition. Trainable prompts based on infrared and ultraviolet images are designed for parameter updates. Throughout the training process, the parameters of the pre-trained large model remain fixed, and only the designed lightweight prompts are updated, significantly reducing the number of training parameters and alleviating the model's dependence on large-scale datasets. The proposed method is compared with several existing methods, and the results demonstrate that this approach can greatly improve the accuracy of power equipment fault recognition, achieving an accuracy of 90.14%. Ablation experiments and visual results further validate the effectiveness of the method. Additionally, the proposed method optimizes only a small number of trainable parameters, ensuring its efficiency.

摘要

针对单谱段图像在电力设备故障识别中的局限性, 提出了一种基于提示学习(prompt learning)的多谱段融合识别方法。为提升大模型对电力设备故障的识别精度, 设计了基于红外图像和紫外图像的可训练提示(prompts), 这些提示作为可训练部分用于模型的参数更新。这种策略很大程度地减少了训练所需的参数量, 且降低了大模型对下游数据量的依赖。利用集成可见光、红外和紫外等谱段的混合成像系统, 对正常和故障电力设备进行了拍摄, 并构建了相应的多谱段数据集, 该数据集经过文本标注后, 可用于大模型的训练。实验结果表明, 所提出的方法可显著提升电力设备故障识别的精度, 平均识别精度达到90.14%。消融实验和可视化结果进一步验证了所提出方法的有效性。此外, 由于所设计的方法只优化了极少数可训练参数, 确保了方法的高效性。