Research on object detection method of infrared porcelain deteriorated insulator based on deep learning

Aiming at the infrared imaging deterioration of porcelain insulator non-contact detection, this research uses a regional network based on the proposed detection: the infrared image deterioration of porcelain insulator was achieved using a Faster R-CNN algorithm. Faster R-CNN has excellent generality and robustness, it can be easily applied to a variety of datasets, and the target categories can be easily adjusted to effectively improve the performance of the test model. To address the limitations of Faster R-CNN in multi-scale detection, this study introduces an enhancement via the Feature Pyramid Network (FPN), which augments the detection accuracy. The FPN enhances the Region Proposal Network (RPN) by integrating the high-resolution features from lower layers with the semantic richness of upper layers, thus bolstering the detection of targets across various scales. Furthermore, the research employs the ResNet101 residual network as a substitute for the VGG16 network, reducing the computational burden of convolutional operations and preserving more informative features of the insulators. This approach facilitates the extraction of detailed features from smaller targets, enhancing detection performance. Experimental comparisons demonstrate that the refined Faster R-CNN algorithm significantly improves the efficacy and precision in identifying insulator defects.