一种改进的DETR输电线通道山火烟雾检测方法

Improved Mountain Fire Smoke in Transmission Line Channel Detection Method for DETR

输电线通道中的火灾会对电力系统的正常运行造成极大的安全隐患,但由于山火烟雾的形状、大小和颜色变化多样导致传统图像方法检测精度较差.为了提高山火检测精度,本文提出一种基于改进DETR的输电线通道山火检测方法.首先,在特征提取阶段加入多尺度特征信息,并利用空洞卷积提高算法对底层特征的感知能力;然后,引入相对位置编码对Transformer模块中的自注意力机制进行改进;其次,利用CIOU对算法的损失函数进行调整;最后,在标注好的输电线通道山火数据集上对改进后算法进行模型训练和测试.实验结果表明,本文所提出的改进后的DETR模型平均精度可达到84.77%,与原始DETR算法相比提高了6.52%,与其它主流目标检测算法对比,本文提出的山火检测模型可有效识别输电线通道中的山火目标并达到较高的检测精度.

The fire in the transmission line channel will cause significant potential safety hazards to the operation of the power system.However,the traditional image processing methods have poor detection accuracy due to the variety of shapes,sizes and colours of mountain fire and smoke.To improve the accuracy of mountain fire detection,a mountain fire detection method in the transmission line channel based on improved DETR is proposed in this paper.Firstly,multi-scale feature information is introduced in the feature extraction stage,and hole convolution is used to enhance the perception ability of the algorithm to the underlying features.Then,relative position-coding is presented to improve the self-attention mechanism in Transformer.Meanwhile,the CIOU-based loss function is utilized.Finally,the proposed method is trained by transfer learning on the labelled mountain fire dataset.The experimental results show that the average precision of the improved DETR model proposed in this paper can reach 84.77%,which is 6.52%higher than that of the original DETR algorithm.Compared with other mainstream target detection algorithms,the mountain fire detection model proposed in this paper can effectively identify mountain fire targets in the transmission line channel and achieve higher detection accuracy.