Addressing elderly health monitoring amid accelerating population aging, this study proposes a non-contact fall detection method using millimeter-wave radar and an improved HDC-RepLKNet network. The TI IWR1843 radar captures human motion echoes, with fused Range-Time Maps (RTM) and Doppler-Time Maps (DTM) providing comprehensive motion characterization. The enhanced network integrates an attention mechanism for refined feature extraction and reduces computational complexity for low-power devices. Tested on a dataset of six actions (three falls, three non-falls), the method achieves 99.92% detection accuracy with strong generalization. This work advances non-contact fall detection and posture recognition, offering a practical solution for elderly care.