Insulator Defect Detection
Electric utilities collect imagery and video to inspect transmission and distribution infrastructure. Utilities use this information to identify infrastructure defects and prioritize maintenance decisions. The ability to collect these data is quickly outpacing the ability to analyze it. Today’s data interpretation solutions rely on human-in-the-loop workflows. This is time consuming, costly, and inspection quality can be subjective. It’s likely some of these inspection tasks can be automated by leveraging machine learning techniques and artificial intelligence.
The Insulator Defect Image Dataset (IDID) consists of labeled high quality images of transmission line insulators. The images have insulator string as the primary subject and parent class. These images contain 3 sub-classes:
1. Flashover damage insulator shell
2. Broken insulator shell
3. Good insulator shell.
The submission should be a csv file with format (imageid, PredictionString)
Here, the image_id is the image name and PredictionString consists of (Class ID, confidence_score, xmin, ymin, xmax, ymax)
The Class IDs are as follows - 0:broken , 1:flashed, 2:good, 3:insulator.
(xyz_imgname1, 2 0.89 232 224 322 512)
(xyz_imgname2, 1 0.94 134 143 458 565)
The evaluation metrics used will be mAp@0.5.