基于DeepLab v3+深度卷积网络的输电导线图像识别方法

输电导线图像识别是电力设备自动巡检过程中的重要环节。针对传统导线检测方法需人工设计目标特征、抗干扰能力差等问题,提出一种基于深度卷积网络（DeepLab v3+）的输电导线图像识别方法。首先,采用DeepLab v3+网络模型,实现导线的初步识别。通过多层卷积自动学习导线特征,并通过解码器结构融合导线低层的细节特征与高层的语义特征,较好地实现导线像素分割。然后,针对识别结果中存在的断裂和伪导线问题,采用改进的最小点对法和长度阈值法进一步精细化处理。最后,采用八方向搜索法提取每一条导线并编号。实验结果表明,所提方法能很好地提取出输电线路图像中的导线。

Image recognition method for transmission line based on the DeepLab v3+ deep convolutional network

Image recognition of transmission line is an important part in the automatic inspection process of power equipment. For the problems of traditional line detection methods that requine manual design of target features and poor generalization ability, an image recognition method for transmission line based on deep convolutional network (DeepLab v3+) is presented. Firstly, the DeepLab v3+ network model is applied to realize the preliminary segmentation of the lines. This model can automatically learn line features by multi-layer convolutions, and merge the low-level detailed features with the high-level semantic features through a decoder structure to improve the accuracy of line pixel segmentation. Secondly, in order to refine the segmentation results, the improved minimum point pair method and length threshold method for removing broken and pseudo lines are proposed. Finally, an eight-direction search method is used to extract and number each line. The experimental result shows that the proposed method can better extract lines in the transmission line image.