Developing an Original Low-Power Neuromorphic Sensor Fusion Architecture for Environmental Monitoring in the Niger Delta

Environmental monitoring in the Niger Delta requires continuous multi-modal sensing for oil spill detection, gas flare monitoring, and water quality assessment, but the power limitations of deployable sensor platforms in remote wetland locations preclude the use of conventional high-power AI inference systems, motivating the development of an original neuromorphic sensor fusion architecture designed for extreme energy efficiency. This study developed an original neuromorphic sensor fusion architecture for environmental monitoring in the Niger Delta, contributing theoretical advances in event-driven multi-modal fusion and a practical neuromorphic design methodology for environmental sensing. The architecture, named EventFusion-Delta, uses a spiking neural network encoder for each sensor modality (acoustic hydrophone, water quality electrochemical array, thermal camera, and spectral gas sensor) that converts continuous sensor streams into sparse spike representations, dramatically reducing computational load compared to frame-based alternatives. An original Temporal Cross-Modal Attention mechanism for spike-based signals was designed, enabling the fusion of asynchronous spike trains from different sensor modalities with different natural time constants, a problem not addressed in existing neuromorphic literature. EventFusion-Delta was implemented in PyNN and benchmarked against equivalent convolutional neural network approaches for oil slick detection and gas flare intensity classification tasks. Energy consumption was 98.7x lower for EventFusion-Delta versus CNN inference at equivalent classification accuracy on sensor data collected from three pilot sites in Rivers and Bayelsa States. The study provides an original neuromorphic architecture contribution directly applicable to Niger Delta environmental protection and recommends HYPREP evaluate the architecture for inclusion in the remediation monitoring technology programme.