基于ConvNeXt和注意力机制的绝缘子自爆故障检测方法

A ConvNeXt and attention mechanism-based method for detecting spontaneous explosive faults in insulators

为了更加准确地识别和定位架空线路绝缘子的自爆故障,保障电力系统安全稳定运行,提出一种基于ConvNeXt和注意力机制的目标检测算法,可用于无人机、巡检机器人等设备拍摄的可见光图像中绝缘子自爆故障检测。首先,使用一种新型卷积神经网络ConvNeXt作为主干网络,使用1∶1∶1∶3的阶段模块数量比例,增强网络对抽象语义特征的提取能力;其次,使用跨阶段局部连接结构,减少网络参数量和计算复杂度,丰富网络梯度连接;最后,引入卷积注意力机制,增强网络对复杂背景中目标区域的感知能力。实验结果表明,改进后的绝缘子自爆故障检测模型的平均精度均值达到97.4%,相比基线YOLOv7提升了1.4%,能够有效实现绝缘子自爆缺陷的检测。

semantic features.Secondly,a cross-stage local connection structure is used to reduce network parameters and computational complexity,thereby enriching network gradient connections.Finally,a convolutional attention mechanism CBAM is introduced to enhance the network′s ability to perceive target areas within complex backgrounds.Experimental results show that the improved insulator spontaneous explosive fault detection model achieves an average precision of 97.4%,an improvement of 1.4%over the baseline YOLOv7,effectively enabling the detection of spontaneous explosive faults in insulators.