HRD-YOLOX based insulator identification and defect detection method for transmission lines
UAV aerial images of insulators have problems such as complex backgrounds, small targets, and obscured targets, resulting in low detection accuracy of insulator targets. In order to solve the above problems, a HRD-YOLOX(Hybrid Attention Mechanisms, Regularization, and Depth Separable Convolution for...
Gespeichert in:
Veröffentlicht in: | IEEE access 2024-01, Vol.12, p.1-1 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | UAV aerial images of insulators have problems such as complex backgrounds, small targets, and obscured targets, resulting in low detection accuracy of insulator targets. In order to solve the above problems, a HRD-YOLOX(Hybrid Attention Mechanisms, Regularization, and Depth Separable Convolution for YOLOX) algorithm suitable for recognizing insulators and detecting defects is proposed based on the YOLOX network. First, the hybrid attention module HAM-CSP(Communication Sequential Processes for Hybrid Attention Mechanisms) is incorporated into the feature extraction network to suppress the interference from the complex background of insulator pictures. Second, the PANet(Path Aggregation Network) structure is replaced with the R-BiFPN(Regularization-Bidirectional Feature Pyramid Network) structure for feature fusion, and an improved regularization module is introduced into the convolutional layer of the BiFPN structure to reduce the overfitting problem of the model and the inference speed of the network. Then, in order to solve the computationally intensive and overfitting phenomenon generated by the detection of small targets such as insulator discharge traces, the DHConv(Deep Separable and Hybrid Attention Mechanisms Convolution) convolution is proposed to be applied to the Conv 3×3 convolutional layer in order to reduce the number of parameters and improve the generalization ability of the model. Finally, the binary cross-entropy function for confidence loss is changed to the Focal Loss function to solve the problem of missed detection and positive and negative sample imbalance due to the image overlap problem The experimental results show that the HRD-YOLOX algorithm improves the average accuracy of insulator detection from 86.23% to 91.34%, and at the same time improves the detection speed from 70.6FPS/S to 82.23FPS/S, which allows real-time detection of insulators, and improves the performance of complex background targets as well as small targets detection. |
---|---|
ISSN: | 2169-3536 2169-3536 |
DOI: | 10.1109/ACCESS.2024.3363430 |