基于CenterNet的输电设备状态缺陷智能识别

CenterNet-Based Intelligent Identification of Transmission Equipment Defects

中文关键词:输电设备缺陷  卷积神经网络  CenterNet  智能识别

keywords:transmission equipment defects  convolutional neural network  CenterNet  intelligent identification

中文摘要:基于无锚框的目标检测算法，提出一种基于CenterNet的输电设备状态缺陷智能化识别方法。通过骨干网络提取图像深层特征信息并生成特征图；通过上采样操作和3个卷积层构成的分支结构，依次输出中心点热力图、中心点偏移量及目标对象的尺寸信息，再借助检测结果解码模块对3类输出结果进行融合处理，最终实现对巡检图像中输电设备缺陷的智能化精准识别。利用建立的输电设备缺陷数据集对模型进行训练和测试，实验结果表明：提出的驱动检测方法，在输电设备缺陷检测任务中展现出优异的检测性能，验证了其实际应用的有效性，且其检测精度完全契合电力智能化巡检的实际需求。该算法的网络架构简洁、支持端到端的模型训练流程，对硬件内存的需求较低，部署难度低，同时能够实现实时在线检测功能。

Abstract:Based on the anchor-free object detection algorithm，this paper proposes an intelligent identification method for transmission equipment defects based on CenterNet. First，the backbone network extracts deep feature information from images and generates feature maps； second，through an upsampling operation and a branch structure composed of three convolutional layers，it sequentially outputs the center point heatmap，center point offset，and target object size information. Subsequently，the detection result decoding module fuses these three types of output results to ultimately achieve intelligent and accurate identification of transmission equipment defects in inspection images. The model is trained and tested using the transmission equipment defect dataset established in this paper. Experimental results show that the proposed method exhibits excellent detection performance in transmission equipment defect detection，verifying its effectiveness in practical applications，and its detection accuracy fully meets the actual requirements of intelligent power inspection. The algorithm features a concise network architecture，supports an end-to-end model training process，has low hardware memory requirements，is easy to deploy，and can achieve real-time online detection.