Implementation of wavelets and artificial neural networks to detection of toxic response behavior of chironomids (Chironomidae: Diptera) for water quality monitoring

Abstract

Movement behavior of Chironomus samoensis larvae was observed in response to the treatments of carbofuran, an anticholinesterase insecticide, at a low concentration (0.1 mg/l) in semi-natural conditions. Two typical movement patterns were selected before and after the treatments, and the variables characterizing movement tracks in two dimensions were analyzed by discrete wavelet transform (DWT) with Daubechie's 4 functions. The variables were selected based on the feature coefficients of DWT and were subsequently used as input for training with the multi-layer perceptron network. The trained network efficiently detected changes in movement patterns before and after the treatments. We demonstrated that the combined use of the wavelets and artificial neural networks would be a useful tool for automatic behavioral monitoring for water quality assessment.