Next-Gen railway asset management with...
summary
Computer vision pipeline at ProRail locates insulation joints from a fleet of inspection images (about 50 million images per campaign), trains a small CNN (three convolutional layers) in TensorFlow, reaches roughly 92% accuracy on the classification task, and covers ~40% of the country so far. Work includes transfer learning experiments, a production data pipeline with prefetch and parallel mapping to feed GPU, and optimizations that make training about ten times faster.
AI-written summary · about
chapters (6)
- 0:00ProRail introduction and use case
- 5:27Computer vision for asset detection
- 12:39Transfer learning for complex tasks
- 17:32Handling large image datasets
- 25:56Optimizing CPU-GPU pipeline
- 27:20Summary and call to action
related talks
41mPractical Aspects of Applying Modern Computer Vision Technologies
36mAI-Powered Real-Time Quality Control
39mPredicting failure & automating defect detection on utility assets
46mML Use Cases at Think Big Analytics
33mRunning Machine Learning in Production - a Journey to Success
24mImage Recognition for safety on the factory floor
59mDeep Neural Networks for Video Applications
11mUsing AI to Accelerate the Smart Green Transition