The behaviour of earthworms is an important indicator for ecological and ecotoxicological studies, providing information on the organisms' responses to environmental stressors such as pollutants and changes in soil composition. However, conventional methods for monitoring soil organisms are problematic due to the opaque nature of the soil matrix. In this study, we present an innovative AIbased system for continuous real-time monitoring of earthworm behaviour in soil environments. The developed system integrates a 2D terrarium experimental setup with a deep convolutional neural network (CNN) to automatically track and analyse earthworm movements. By recording and processing image sequences, the system quantifies key behavioural endpoints such as total tunnel length, re-exploration rate and total distance travelled, providing valuable metrics for understanding earthworm responses. In addition, new parameters relating to movement patterns and tunnelling efficiency have been introduced to further enrich the dataset. To test the performance of the system, we conducted three different experiments. First, a modified avoidance test was performed with H3BO3 as reference contaminant to evaluate the system's ability to detect behavioural responses under proven conditions. In the second series of experiments, we tested the effects of various commonly used pesticide formulations, including organophosphates, carbamates and triazoles, mixed uniformly into the soil. Finally, we investigated the performance of the system in different soil compositions by varying the ratio of quartz sand, kaolin clay and Sphagnum peat to simulate different environmental conditions. In all three experiments, the AI-based monitoring system showed consistent accuracy, reliability and improved precision compared to conventional methods. The proposed AI-based monitoring approach represents a breakthrough in soil health studies. It provides a cost-effective, rapid and objective tool to assess the ecological impact of contaminants on soil organisms. By continuously and unobtrusively monitoring the behaviour of earthworms, this system provides a robust platform for ecological monitoring and pesticide risk assessment and advances soil health research.