engleski

Earthworm avoidance behavior quantification using artificial neural networks

Avoidance behavior of earthworms represents a valuable endpoint in ecotoxicological studies examining the effects of sub-lethal concentrations of pollutants on soil-dwelling fauna. Since it was first performed by Yeardley (1996), the earthworm avoidance test was proven to give ecologically relevant insights into modes of action of various pollutants. Since behavioral responses of test organisms are highly variable over time, development of techniques that would allow continuous monitoring of avoidance behavior are of great importance. In this research a novel approach is presented towards developing a low-cost, automated monitoring method for continuous examination of earthworm behavior within avoidance tests. The suggested system implements a 2D terrarium, a tool widely used in earthworm behavioral studies, as a compartment in which the avoidance test is performed. Automatization of avoidance behavior detection analysis is achieved using a deep convolutional neural network model, constructed and trained in Keras, that gives precise predictions of earthworm locations in images of both sides of the terrarium during the test period. To quantify the motion of the experimental organisms in 2D space, the area of the terrarium is divided into multiple segments. Performance of the presented system was tested in avoidance tests performed in glass terrarium filled with clean artificial soil in one half and artificial soil spiked with in oil mill waste in the other half. Predictions of earthworm location probabilities over the image-pairs show high level of overlap, complementing each other and increasing certainty of automatic earthworm location determination. The first period of the experiment is characterized by equal presence of earthworms in both halves of the terrarium, with a gradual shift from the bottom to the top layers. Thereafter, a transition to the unpolluted (control) half is observed. This period of the experiment is characterized by the localization of earthworms in the bottom half of the control soil. The presented ANN-based approach proved to be useful in continuous monitoring of earthworm behavior. Due to its low cost and simple design, it is an easily reproducible setup that gives a significant acceleration to analysis and interpretation of data resulting from behavioral studies, with a potential of becoming an ultimate tool in research of primary modes of action of various pollutants present in the environment. Keywords: avoidance behavior, continuous monitoring, U-net, automatization

avoidance behavior ; continuous monitoring ; U-net ; automatization

nije evidentirano