The increasing prevalence of falls among the elderly highlights the urgent need for advanced monitoring systems capable of distinguishing high-risk impacts from routine movements. However, existing solutions often struggle with challenges such as reliance on external power sources, privacy concerns, limited sensing coverage, and insufficient sensitivity, flexibility, or contextual intelligence for reliable detection. To address these limitations, we report a bio-inspired, fully flexible triboelectric nanogenerator (TENG) sensor integrated with deep learning. The sensor leverages a biomimetic design (a micro-structured Ecoflex layer patterned with Osmanthus fragrans leaves) paired with a conductive sponge electrode, to achieve robust mechanical properties (1049.21% strain tolerance and ~ 16.8k cycles durability) and high sensitivity (42.7 V/g). By encoding key biomechanical parameters (impact location, force, and frequency) into distinct electrical signals, the TENG generates site-specific profiles when adhered to injury-prone areas such as the side waist, shoulder, knee, and palm. These signals are processed by a 1D convolutional neural network (1D-CNN) enhanced with Squeeze-and-Excitation modules, achieving a classification accuracy of 91%. The system prioritizes high-severity impacts (e.g., hip (side waist) strikes) through context-aware analysis, enhancing emergency response precision while minimizing false alarms. This work bridges material innovation, biomechanical sensing, and adaptive artificial intelligence (AI), offering a scalable, wearable solution for real-time fall detection and injury risk assessment in elderly care.