基于密集连接网络的航拍绝缘子旋转目标精准定位方法

为了实现架空线路巡检时绝缘子的精准定位和检测，提出了一种基于Dense-Block密集连接块与旋转框改进YOLOv5的绝缘子检测模型。该模型针对绝缘子长宽比较大和方向多变的特点，提出利用长边定义法为检测框增加角度信息，实现目标旋转框检测，有效提升绝缘子检测和定位的效果。同时为了增强特征的重新利用和传播，利用Dense-Block对模型中的残差模块进行改进，构建YOLOv5-dense检测模型。最后为了使YOLOv5-dense模型能够更加关注有效的特征信息，在主干网络尾部加入SimAM注意力模块对模型进行改进。实验之前，利用Retinex算法对输入绝缘子图像进行增强。实验结果表明，相较于原始YOLOv5算法，所提算法在平均准确率和每秒处理帧数方面都有提高。除此之外，与水平框检测算法相比，所提算法去除了检测结果中大量冗余的背景信息，实现了绝缘子区域更加精准的定位。

Accurate positioning method of insulator rotating target in aerial photography based on dense connection network

In order to realize the accurate positioning and detection effect of insulators during circuit inspection, this paper proposes an improved YOLOv5 insulator detection model based on Dense-Block and rotating frame. Aiming at the characteristics of large length width ratio and changeable direction of insulator, this model proposes to use the long side definition method to add angle information to the detection frame, realize the target rotation frame detection, and effectively improve the effect of insulator detection and positioning. At the same time, in order to enhance the reuse and propagation of features, this paper uses dense block to improve the residual module in the model and build YOLOv5-dense detection model. Finally, in order to enable the YOLOv5-dense model to pay more attention to effective feature information, a SimAM attention module is added at the end of the backbone network to improve the model. Before the experiment, Retinex algorithm is used to enhance the input insulator image. The experimental results show that compared to the original YOLOv5 algorithm, the algorithm proposed has improved average accuracy and processing frames per second. In addition, compared with the horizontal frame detection algorithm, this algorithm removes a large amount of redundant background information in the detection results, and realizes more accurate positioning of the insulator area.