engleski

Observing earthworm behavior using deep learning

Earthworm behavior represents a valuable endpoint in ecological and ecotoxicological studies, providing insight into individual- and population-level responses of organisms to ecological factors and xenobiotics. Difficulties in observing activities of soil-dwelling organisms come from the soil matrix surrounding the animals. We present an automated system for continuous monitoring of earthworm behavior directly in the soil. Performances of the developed system are tested in an avoidance test set-up. The developed system includes a 2D terrarium serving as an experimental compartment and a deep convolutional neural network model for continuous monitoring of earthworm behavior. Quantification of activity of organisms is based on the predictions of the neural network model made from image sequences captured during the exposure. Performance of the system was validated by comparison with results of the standard avoidance test, using H3BO3, a standard pollutant. The presented system represents a simple, cost-effective, fast, accurate and objective tool for continuous earthworm behavior monitoring in ecological and ecotoxicological studies. Source code and training data of this system are made freely available on GitHub.

Earthworm ; Behavior ; Burrow ; Deep learning ; Computer vision

nije evidentirano