1) Combined FRCNN with custom pose estimation to identify defects in insulators on transmission lines.
2) Developed an image segmentation module for the detection of missing nuts and bolts in transmission towers. Used a classifier to eliminate 90% of the wrong predictions.
3) Trained a customized variant of Google DeepLabs V3+ for segmentation of metal from images of transmission towers for rust detection.
4) Created a safety tool by training a RetinaNet model on construction site images for generating safety alerts in unsafe situations.
5) Created an automated transmission tower detection module for massive GeoTIFF satellite images to identify and geotag towers for future planning and management.