基于改进SSD的变压器套管红外图像油位智能识别方法

为解决现有基于红外图像识别变压器套管油位存在的过于依赖温度信息、人工处理效率低下等问题,文中结合目标检测技术提出了一种基于改进单次检测器(SSD)的套管智能油位识别方法。通过引入SSD目标检测方法,检测红外图像中的套管区域,加入损失函数以改进SSD算法从而提高套管检测准确率,并进一步通过简单线性迭代聚类(SLIC)的应用实现了不依赖红外图像温度信息的油位检测。对比文中提出的基于红外图像的油位识别算法检测结果与人工油位检测结果,表明文中提出的算法不仅在效率上领先于传统的温度检测方式,且其误差较小,仅为0.08%。对比结果验证了所提算法在保证检测精度的情况下可大幅度提高检测效率,有效提升套管故障诊断效率和智能化水平。

Intelligent oil level recognition of transformer bushing infrared image based on improved SSD algorithm

Intelligent oil level recognition based on the infrared image of the bushing is faced with the problems of the complicated background of collected infrared images and the difficulty of obtaining temperature information from infrared images in batches. To solve the problem of complex infrared image background, the SSD object detection algorithm is employed to detect the bushing area in the infrared image. At the same time, in order to solve the over-fitting problem of the bushing infrared datasets, the center loss function is added to improve the SSD accuracy. Based on the SLIC (Simple Linear Iterative Clustering) pre-procession, the oil level detection algorithm is designed without relying on the temperature information from the infrared images. Finally, the proposed algorithm is compared with the oil level recognition algorithm based on temperature. The results show that the improved SSD can improve the accuracy of object detection, and the relative error between the oil level recognition algorithm and temperature oil level detection is 0.08%. It proves that the proposed improved SSD algorithm can be effectively applied to the intelligent identification of oil level based on the infrared image of the bushings, and detect bushings in the complex background from infrared image, which can greatly improve the efficiency and intelligent level of bushing fault diagnosis.