Defect Detection of Lightweight Glass Insulator Based on Improved YOLOv4
Aiming at the problem of poor real-time detection of glass insulators and their defects during the in-spection process of power UAV,a defect detection model of lightweight glass insulators based on improved YOLOv4 was proposed.First,the MobileNetV3-Large backbone network was improved and used as the backbone network of YOLOv4.Secondly,a lightweight convolution method was proposed,which can greatly reduce the amount of calculation and improves the inference speed under the premise of ensuring high accuracy.Then the ReLU6 function was used as the activation function to improve the performance of the model.Next,the Inception-Resnet structure was introduced into the feature fusion module to obtain feature maps that are more suitable for detection.Finally,a multi-stage transfer learning ideological training model was adopted to improve the training efficiency.Experiments show that compared with the YOLOv4 model,the parameters of the model in this paper are reduced by 198.39M,the MAP is increased by 11.31%,and the detection speed is about 2 times and 5.8 times that of the original on GPU and CPU devices.and real-time detection of its defects.
NETL
NSTL
万方数据
