The Application of Nueral Networks to Size Analysis of Material on Conveyors

The underlying theory behind the construction of an on-line size analysis system is described together with an introduction to the Artificial Intelligent techniques that it is founded on. Previous systems based on more traditional technology such as edge detection and dilation algorithms have been restricted by the significant processing power they require, this has resulted in the production of systems with either reduced functionality or severe hardware requirements. A neural network approach offered a new perspective to the problem of size analysis; the ability of a network to learn and recognise subtle patterns in training data before on-line processing is initiated opened up the possibility for the production of a fully functioning practical system. The format of the training data had to be carefully analysed prior to its implementation in a neural network as the capability of a neural network solution depends as much upon a suitable data preparation as it does on the system architecture. To facilitate experimentation with various profile quantification techniques such as fourier and fractal analysis an Image Profile Analysis System was constructed (IPAS) in software on a personal computer. The IPAS system was produced with the aim of pre-processing image data to create data sets for experimenting with external neural network packages. The system allows the analysis of both real images obtained by image capture and simulated images created by an in-built in simulator. A measurement comparison scheme is also included together with a case study concerned with the frequency analysis of simulated images. The power of neural networks combined with the data compression provided by profiling removes the need for direct measurement of individual particles. This has obvious implications on the level of hardware necessary, and ultimately the expense of a stand alone system. The extrusive nature of data collection via an image analysis system will also permit simple installation of a shipped system; eliminating any requirement for interference with existing machinery.