Abstract
This paper reports on three methods of classifying the spectral data from an optical fibre based sensor system as used in the food industry. The first method uses a feed-forward back-propagation Artificial Neural Network; the second method involves using Kohonen Self-Organising Maps while the third method is k-Nearest Neighbour analysis. The sensor monitors the food colour online as the food cooks by examining the reflected light from both the surface and the core of the product. The combination of using Principal Component Analysis and Backpropagation Neural Networks has been successfully investigated previously. In this paper, results obtained using all three classifiers are presented and compared. The Principal Components used to train each classifier are evaluated from data that generate a "colourscale" comprising six colour classifications. This scale has been developed to allow several products of similar colour to be tested using a single network that had been trained using the colourscale. The results presented show that both the neural network and the Self-Organising Map approach perform comparably, while the k-NN method tested under-performs the other two.
Original language | English |
---|---|
Article number | 94 |
Pages (from-to) | 706-713 |
Number of pages | 8 |
Journal | Proceedings of SPIE - The International Society for Optical Engineering |
Volume | 5826 |
DOIs | |
Publication status | Published - 2005 |
Event | Opto-Ireland 2005: Optical Sensing and Spectroscopy - Dublin, Ireland Duration: 4 Apr 2005 → 6 Apr 2005 |
Keywords
- Artificial Neural Network
- k-NN
- Optical Fibre sensor
- Self-Organising Map