基于改进的CenterNet变电站设备红外温度检测方法

Infrared temperature detection system for CenterNet substation equipment based on improved

摘要:

红外检测能够检测变电站电力设备温度异常，降低安全事故发生的概率，因此，提出一种基于改进的CenterNet目标检测算法模型CenterNet_PRO。该算法采用了ShuffleNet V1/V2作为骨干网络、引入了FPN来提取多尺度特征，为了克服不同尺度目标检测的难点、增加旋转角度回归分支，用于预测目标的旋转角度以及改进的IoU Loss进行优化，进一步提高模型检测速度和准确率。通过阈值分割法提取电力设备表面温度并分析计算，设计制定电力设备温度缺陷判断规范、温度警告阈值，根据该规范即可判断电力设备的相关缺陷。实验结果表明，改进的CenterNet模型平均精度达到了90%，相比于传统的CenterNet模型，平均精度提高了1.3个百分点，可以满足实际变电站场景下对电力设备红外检测的高要求。

Abstract:

Infrared detection can detect temperature anomalies in substation power equipment and reduce the probability of safety accidents. Therefore, an improved CenterNet object detection algorithm model, CenterNet, is proposed_ PRO. This algorithm adopts ShuffleNet V1/V2 as the backbone network and introduces FPN to extract multi-scale features. In order to overcome the difficulties of target detection at different scales, increase rotation angle regression branches, predict the rotation angle of the target, and optimize with improved IoU Loss, further improving the model detection speed and accuracy. The surface temperature of power equipment is extracted by the threshold segmentation method and analyzed and calculated, and the temperature defect judgment specification and temperature warning threshold of power equipment are designed and formulated, and the related defects of power equipment can be judged according to the specification. The experimental results show that the average accuracy of the improved CenterNet model reaches 90%. Compared with the traditional CenterNet model, the average accuracy is improved by 1.3 percentage points, which can meet the high requirements for infrared detection of power equipment in the actual substation scenario.