基于图像切片的移动端输电线路鸟类检测算法研究

Research on Mobile Bird Detection Algorithm of Transmission Line Based on Image Slice

输电线路作为电网的基础组件,其故障是影响电网稳定运行的主因,其中鸟类是输电线路的主要安全隐患。为了实现低耗能、高精度的驱鸟,论文提出一种基于图像切片的移动端鸟类检测算法,同时为减少图片传输的网络时延及避免因网络中断造成的数据丢失,鸟类检测在移动终端实现。但是移动终端的计算性能和存储性能较低,无法直接运行基于深度学习的目标检测算法,因此论文将图像进行切片把目标检测问题转换为图像分类问题,采用ResNet18对切片进行鸟类识别。此外,为进一步减少无鸟图像的识别时间,在对每一个切片进行识别之前,先计算实时采集图像和不包含鸟的模板图像的差值,只有差值较大时才进行识别。实验证明,论文提出的模型在精度和速度上都可以在移动端上满足输电线路鸟类检测任务的要求,具有推广意义。

As the basic component of the power grid,the fault of the transmission line is the main factor affecting the stable op⁃eration of the power grid,among which the bird is the main safety hazard of the transmission line.In order to realize low energy con⁃sumption and high precision bird driving,this paper proposes an algorithm of mobile terminal bird detection based on image slices.Meanwhile,in order to reduce network time delay of image transmission and avoid data loss caused by network interruption,bird de⁃tection is implemented in mobile terminal.However,the computational performance and storage performance of the mobile terminal are relatively low,so it is impossible to directly run the target detection algorithm based on deep learning.Therefore,this paper con⁃verts the problem of target detection into the problem of image classification through image slicing,and uses ResNet18 to identify birds by slicing.In addition,in order to further reduce the recognition time of birdless images,the difference between the real-time collected images and the template images without birds is calculated before the recognition of each slice,and the recognition is car⁃ried out only when the difference is large.Experimental results show that the model proposed in this paper can meet the require⁃ments of bird detection in transmission line in terms of accuracy and speed.