Michael Free from BT gave an interesting presentation at Cogex yesterday about using convolutional neural networks to analyse images of telephone exchanges. Traditional types of neural network image recognition perform poorly on images with shadows and wires in different locations. Using a convolutional neural network together with deconvolution gives significant improvements. This enables field engineers to capture images on location of the exchange which are then automatically processed for asset management data.
In machine learning, a convolutional neural network is a class of deep, feed-forward artificial neural networks, most commonly applied to analyzing visual imagery.