An Original Framework for Neuromorphic Computing Architectures in Energy-Constrained Edge AI Applications for Nigerian Agriculture

Neuromorphic computing architectures inspired by biological neural information processing offer theoretical advantages in energy efficiency and event-driven computation that could enable practical AI deployment on solar-powered agricultural edge nodes in Nigeria, where energy scarcity constrains conventional GPU or NPU inference. This study developed an original neuromorphic computing framework for energy-constrained agricultural edge AI applications, grounded in a combination of theoretical analysis and empirical evaluation. A systematic review of neuromorphic hardware (Intel Loihi 2, IBM TrueNorth, SpiNNaker 2) and spiking neural network training approaches from 2020 to 2024 was conducted, identifying that no study had evaluated neuromorphic approaches for precision agriculture tasks under energy budgets compatible with 20W solar-powered edge nodes typical of Nigerian farm deployments. The original Spiking Agricultural Neural Network (SANN) architecture was designed for three agricultural inference tasks: crop disease classification, soil moisture state estimation, and pest detection from trap camera images. SANN was implemented in PyNN for simulation on SpiNNaker 2 and converted to a neuromorphic-aware quantised format for Intel Loihi 2. Energy consumption benchmarking showed SANN achieving crop disease classification at 0.42 mJ per inference compared to 18.7 mJ for an equivalent MobileNetV2 INT8 model on Cortex-M7, a 44.5x efficiency improvement. Accuracy on the cassava disease classification task was 81.4 percent versus 87.4 percent for the non-neuromorphic baseline, representing a 6-percentage-point accuracy-efficiency trade-off that favoured neuromorphic deployment for energy-critical applications. The framework constitutes an original contribution to the emerging field of neuromorphic agricultural AI.

Keywords: neuromorphic computing, spiking neural network, agricultural AI, edge computing, energy efficiency