Development of an artificial neural network for automated age estimation
Current age determination methods, even when aided by image analysis software still depend on interpretation by an experienced "reader". The process of ageing is also laborious, time consuming and hence, relatively expensive. For production ageing, where there is an ongoing requirement for age estimates, there is a problem of consistency of interpretation. At present, when readers change, there is a substantial training and verification period needed to ensure that the new reader is interpreting otolith structure in a consistent and correct manner. Automatic ageing would have the primary advantage of being a far more objective method than is possible with even the best training, reducing discrepancies both between readers and organisations. This factor will increase the precision of estimates and therefore provide greater confidence for the stock assessment process. Benefits associated with the development of this technique also include the reduced sample processing time which would increase the number of samples able to be processed and hence, reduce the cost. The pilot project which has been completed has demonstrated the potential for artificial neural networks to objectively and consistently classify samples of some species. With refinements of the system, it should be applicable to any species for which production ageing is required.
1. Compare the effect of different forms of data input on the performance of an ANN model for automatic ageing.
2. Compare the effect of different forms of ANN models on their performance.
3. Develop a protocol for the application of an ANN model to the process of automatic ageing.
Principal Investigator: Simon Robertson and Alexander Morison