This paper develops a lightweight AkidaNet model to identify wild radish weeds in canola crops at four early growth stages, achieving 99.73% classification accuracy on the bccr-segset dataset and outperforming several standard CNN architectures. The model is then quantised and converted to a Spiking Neural Network for deployment on BrainChip’s Akida neuromorphic hardware, retaining comparable accuracy while delivering low-latency, low-power inference verified via Grad-CAM.