Machine Learning for Object Detection & Automated Image Extraction of Electric Utility Assets
Machine Learning for Object Detection & Automated Image Extraction of Electric Utility Assets
Business Overview
A major electric utility required an automated method to extract clear and accurate images of utility poles and attached equipment captured through 360° panoramic camera systems (Ladybug camera). This solution aimed to eliminate manual review, accelerate field asset documentation, and improve data accuracy for asset inventory and maintenance planning.
Challenge
The client relied on continuous panoramic photo streams captured from vehicles traveling along utility corridors. Manual identification and extraction of utility pole images from these large datasets was:
- Time-consuming and labor-intensive
- Prone to human error and inconsistency
- Difficult due to factors like lighting conditions, sun glare, noise etc
- Additionally, the output needed quality validation to avoid storing blurred, obstructed, or irrelevant images.
Solution
The team of GIS experts at NeST Digital implemented a Machine Learning–powered automated image extraction system designed specifically for electric utility asset recognition.
Key Capabilities Delivered
- Automated Object Detection & Extraction
- Application analyzed 360° video/photo stream files using integrated vehicle GPS and asset coordinate data
- For each utility pole along the network, the system extracted relevant cropped images automatically
- Deep Learning Classification & Validation
- Prepared ML training dataset using utility pole images
- Developed an image classification model using LeNet-5 Neural Network architecture
- Implemented classification and validation using TensorFlowSharp
- Image Pre-Processing & Quality Filtering
- Grayscale tuning, histogram equalization, normalization, and dropout processing
- Sun-glare detection logic to exclude obstructed images
- Artifact detection and automated rejection
- Scalable Client–Server Architecture
- Image classification tool deployed as a distributed system to support large-scale asset inventories
Impact
- 90%+ reduction in manual effort through automated identification and extraction
- Improved data accuracy & consistency using machine learning validation and automated quality control
- Faster asset inventory cycles, enabling large-scale field data processing
- High scalability and reusability across multiple infrastructure asset types and industries
- Better predictive maintenance data integrated into enterprise systems
SHARE
