DETECTION AND CLASSIFICATION OF OVERHEAD CONDUCTORS AND PIN INSULATORS OF DISTRIBUTION LINES USING MASK-RCNN

ABSTRACT

To ensure stability and efficiency of power supply, the Electric grid is routinely inspected to check for defects that may ultimately lead to faults. Faults in the electric grid can occur as a result of defects in power line components, hence the need for timely detection of defects as a preventive and corrective measure. Manual inspection of power line components is slow and associated with high costs of hiring personnel, and does not ensure the safety of inspection personnel, Hence, the need for the development of an automatic defect detection system which is faster, reduces the costs and risks associated with manual inspection, and in turn increases the Efficiency of the Electric grid. This project is about the development of a system capable of detecting overhead conductors and pin insulators of distribution lines, and identifying their states (defective or normal). This work presents a novel approach for detection and classification of overhead conductors and pin insulators of distribution lines, by exploiting the state-of-the-art Mask R-CNN (Mask region based convolutional neural network). The method consists of five steps: (I) Data collection, (II) Data preprocessing (III) polygonal annotations of images, (IV) Training Mask RCNN models using transfer learning and hyperparameter fine tuning, with ResNet-101 as backbone CNN, and (V) Evaluating the performance of the algorithm in terms of mean average precision and recall at different IoU thresholds. The developed Mask-RCNN based method produced very encouraging results, achieving mean average precision of 0.99 for insulators with a recall of 1.00, and for the conductors, mean average precision and recall values of 0.91 and 0.93 respectively. The method developed in this work can be potentially useful for power distribution companies where routine inspection is key to maintaining efficiency in power supply.