The growing demand for intelligent and sustainable agriculture has accelerated the adoption of deep learning–based solutions for crop monitoring and disease detection. However, deploying these models on resource-constrained embedded platforms remains challenging due to limited memory and computational power. In this work, we optimize YOLOv8n, a lightweight object detection model, for real-time plant disease recognition on edge devices. By applying structured pruning, knowledge distillation, and quantization, we reduce the model size by 75%, the FLOPs by 17.56% and cut inference time by 50%. It achieved a mAP50 of 94.10% on the test dataset and reached an inference speed of up to 185.93 FPS on the Jetson Orin Nano. The optimized model was deployed on Raspberry Pi, NVIDIA Jetson, and Android smartphones, confirming its suitability for real-time applications on resource-constrained platforms. This study demonstrates, for the first time, the combined use of pruning, distillation, and quantization for YOLOv8n in precision agriculture, paving the way for scalable AI solutions that can transform crop management and disease prevention in the field.