Detection Of Fault Insulator Based On Improved Convolution Neural Network

An insulator is an essential component of a transmission line, serving to prevent the leakage of electricity flow from the conductors into the ground. It accomplishes this by creating a barrier between the conductors and the supporting structure. The insulator's atomic structure consists of electrons that are strongly bound and exhibit limited mobility. Researchers have researched a variety of methods for detecting insulators through the use of image processing in previous studies. The majority of contemporary detection systems use classifiers for this purpose. These methods use a classifier trained on a training set of images to recognise an object in a test image, despite the fact that there are a few drawbacks in terms of detection precision and speed. This thesis proposes a method for constructing a hybrid YOLOv5-Resnet50 system, with Resnet50 serving as the backbone of the YOLOv5 architecture. Using a hybrid of alternating and altering the backbone of the YOlOv5s structure, the proposed method achieves an accuracy of 99.0 ± 0.233% and a training time of 25 minutes for a set of 1,000 insulator images. This proposed method has the potential to aid in the inspection of high-up insulators, and it aims to reduce the manpower required to perform this task, which is one of the most dangerous and has a high fatality rate due to its high-voltage field and high-altitude placement. Future plans include expanding the dataset size in order to enhance the system further. Next is utilising a very high-end Specification of equipment by utilising a very excellent GPU and CPU to train the data more effectively