基于改进CNN的电力设备红外目标检测技术

Infrared target detection technology of power equipment based on improved CNN

电力设备红外图像背景较为复杂、设备种类较多,使得电力设备红外目标检测效果较差。为此,提出基于改进CNN的电力设备红外目标检测技术研究。通过联合中值滤波与高斯滤波,预处理电力设备红外图像。以此为基础,采用ResNet与FPN改进CNN,获取电力设备红外图像特征信息。通过常数分割模型训练随机森林分类器,将特征信息输入至训练后的分类器中,获得电力设备红外目标检测结果。实验结果显示,采用所提方法对电力设备红外目标检测结果与实际检测结果保持一致,帧率最大值为14.1 FPS,表明所提方法应用性能良好。

Due to the complex background and variety of devices in the infrared image of power equipment,the detection effect of infrared targets in power equipment is poor.Therefore,a research on infrared target detection technology of power equipment based on improved CNN is proposed.The infrared image of power equipment is preprocessed by joint median filter and Gaussian filter.Based on this,ResNet and FPN are used to improve CNN to obtain infrared image feature information of power equipment.The random forest classifier is trained by constant segmentation model,and input the feature information into the trained classifier to obtain the infrared target detection results of power equipment.The experimental results show that the infrared target detection results of power equipment using the proposed method are consistent with the actual detection results,and the maximum frame rate is 14.1 FPS,indicating that the proposed method has good application performance.