基于改进CenterNet的电力设备红外图像识别

Infrared Image Recognition of Power Equipment Based on Improved CenterNet

由于受到变电站复杂背景的影响,所以传统目标检测算法无法准确识别和定位电力设备。提出了一种基于改进中心点网络(CenterNet)的电力设备红外图像识别模型。首先,针对复杂环境下目标特征信息不足的问题,使用特征提取能力更强的ResNeXt50作为CenterNet的主干网络,在保持原模型参数量不变的同时增加了网络宽度,使其能获取更加丰富的特征信息,从而提升检测的精度;然后,通过在预测层加入通道注意力机制来提升模型对重要目标的关注度,同时抑制无关信息干扰,保证了模型检测的鲁棒性;最后,为证明模型的有效性,在自制数据集上进行了实验验证。结果表明,改进后模型的均值平均准确率可达93.7%,相比原始模型提升了2.1%,优于现有的几种主流检测模型,有效提升了变电站复杂环境下电力设备红外图像识别的精度。

Traditional target detection algorithms are affected by the complex background of substations and cannot accurately recognize and locate the electrical equipment.Therefore,this paper proposes an infrared image recognition model for the electrical equipment based on an improved CenterNet algorithm.Firstly,to address the problem of insufficient target feature information in complex environments,the ResNeXt50 with stronger feature extraction ability is used as the backbone network of CenterNet.While keeping the original model parameters unchanged,the network width is increased to obtain more abundant feature information,thereby the detection accuracy is compared.Secondly,the channel attention mechanism is added to the prediction layer to enhance the model’s attention to important targets while suppressing irrelevant information interference,ensuring the robustness of model detection.Finally,to prove the effectiveness of the model,verification experiments were conducted on a self-made dataset.The results show that the mean average accuracy of the improved model reached 93.7%,an increase of 2.1%compared to the original model,which is better than several existing mainstream detection models.This method effectively improves the recognition accuracy of electrical equipment infrared images in complex substation environments.