Using artificial intelligence to detect fish breathing stress in remote automated biosensing for coordinated watershed monitoring networks

摘要

Analog signals generated by fish gill ventilatory events were digitized from rock bass (Ambloplites rupestris, Raf.) in near, real-time. Signals were processed using Fast Fourier Transform (FFT) to establish continuous frequency-domain records (i.e., Power Spectrum Densities [PSD]) representing both qualitative and quantitative features of the data. Employing an automated biosensing system under laboratory conditions PSD features for both ventilatory events and water quality vectors (i.e., temperature) were used as input to train an artificial neural network (ANN) classifier to establish pattern recognition of ventilatory responses under ambient and temperature stressed states. Fish gill ventilatory signatures were generated by first training an ANN classifier for fish maintained under ambient states to recognize non-stress events and then to establish critical thresholds of thermal stress for the automated biosensing system. Once established we tested the null hypothesis that there would be no difference between non-stress and thermal stressed fish gill ventilatory signatures employing a probabilistic approach to pattern recognition or Baye's classifier, making a comparison of ANN and Baye's classification techniques. When implemented as remote platforms in proposed coordinated watershed monitoring networks where a significant number of sensed fish exceed critical thresholds of safety for specified durations, a toxic episode may be declared, triggering a series of feedback-control options designed to protect the water resource.