Development of an intelligent reasoning system to distinguish hunger states in Rainbow trout (Oncorhynchus mykiss)

In aquaculture it is inherently difficult to monitor animals as they are underwater. Furthermore, feed costs account for 50–60% of production costs. Recently, telemetry technology has advanced such that physiology and behaviour of fish can be monitored in real time, providing the incentive for automated monitoring systems. As muscles become activated, they generate electric signals, known as electromyograms (EMG's). These EMG's can be measured in fish with the use of a radio-transmitter, an EMG tag. Previous studies have demonstrated collection of telemetric physiological and behavioural data on feeding fish; however, automation of the classifying process has not been attempted. The following paper details the use of supervised pattern recognition in successfully classifying (average success rates 85%) fed and fasted fish. In addition, the success rate of quadratic and support vector machine (SVM) classifiers, different feature sets and data sets is examined.