基于深度学习SSD的变电站设备故障视频识别

Fault video recognition of substation equipment based on deep learning SSD

变电站设备故障视频识别结果受随时间呈动态性改变视频图像影响,导致识别精度较低。为此提出了基于深度学习SSD的变电站设备故障视频识别方法。设计固定大小的滑动时间窗口,采集并标注变电站设备故障图像;构建基于深度学习SSD识别框架,计算卷积层感受野,获取故障设备及其周围环境信息尺度信息;计算各个特征层像素上默认框大小,结合目标损失函数,改善识别框位置可信度,设计识别规则,实现变电站设备故障视频的识别。实验结果证明,该方法能够保证识别框与真实框重合程度在0.8以上,说明使用该方法能够相对精准地识别变电站设备故障。

The video recognition results of substation equipment faults are affected by the dynamic changes in video images over time,resulting in low recognition accuracy.A fault video recognition method for substation equipment based on deep learning SSD is proposed for this purpose.Design a fixed sliding time window,collect and annotate fault images of substation equipment.Construct a deep learning SSD recognition framework,calculate the receptive field of convolutional layers,and obtain scale information of faulty devices and their surrounding environment.Calculate the default box size on each feature layer pixel,combined with the target loss function,improve the credibility of the recognition box position,design recognition rules,and achieve the recognition of substation equipment fault videos.The experimental results show that this method can ensure that the degree of overlap between the recognition box and the real box is above 0.8,indicating that the use of this method can relatively accurately identify substation equipment faults.