Table 2.

Hardware-aware optimization strategy adopted for embedded deployment.

Platform	Backend	Optimization	Precision
Jetson Nano	TensorRT	Layer fusion, kernel tuning, GPU memory optimization	FP16/INT8
Raspberry Pi 5	ONNX Runtime	Graph optimization, operator fusion, multi-threading	FP32
Raspberry Pi 5 + Intel VPU	OpenVINO IR	MYRIAD plugin, CPU–VPU execution	FP16
Raspberry Pi 5 + Hailo-8 NPU	HailoRT/HEF	INT8 quantization, graph partitioning, kernel fusion	INT8