Artificial Neural Networks

Artificial neural networks can be trained to perform specific tasks by showing them how to respond to particular examples. When this technique is combined with machine vision technology, powerful systems result that can learn visual tasks such as inspection and control.

Neural Networks

Artificial neural networks are computer models of simple brains. An artificial neural network contains many simple processing elements that, like natural neurons, learn to solve particular tasks.

There are many types of artificial neural networks, designed to mimic different capabilities of natural neural networks. Some artificial neural networks are designed purely to solve particular types of problems and bear little or no resemblance to natural neurons.

Feed-forward ANNs

A feed-forward artificial neural network is designed to look at an input pattern and respond with a particular output. It is trained to perform a task by presenting it with example patterns of inputs and outputs. The network is trained to reproduce the example outputs when presented with the corresponding inputs.  After training, the network can perform the task for new inputs that it has never seen before.

The network consists of nodes that perform simple calculations, and connecting weights between the nodes. Training adjusts the connecting weights so that the network produces the correct responses to the example inputs.

Applications

Feed-forward neural networks have been successfully applied to a wide variety of tasks including driving a vehicle, detecting abnormal cells in images, and financial prediction.

Neural Networks and Machine Vision

The incorporation of Artificial Neural Networks in Machine Vision Systems provides systems that can be trained by showing them examples of the visual properties they are to identify. Hamey Vision Systems have used this powerful combination of techniques to develop a biscuit bake inspection system in collaboration with Access Macquarie Ltd and the CRC for International Food Manufacture and Packaging Science. The system can be trained by showing it examples of biscuits with different bake levels. This makes the system easy to use and flexible, able to work with a wide variety of products.