Shack-Hartmann wavefront sensor (SHWFS) faces an intrinsic trade-off between detection sensitivity and microlens spatial sampling. We present an 'Ultra-Compact Sensing—AI Decoding' solution integrating Four-Quadrant SHWFS (FQ-SHWFS) with Quadrant Intensity Encoding Network (QIE-Net). To our knowledge, it is the first scheme achieving direct high-precision wavefront reconstruction under extremely low spatial sampling resolution. Simulations and experiments demonstrate that the proposed method achieves a residual wavefront RMS of approximately 0.01μm, with a limiting magnitude improvement of over 4mag compared to conventional SHWFS. Knowledge distillation optimizes computational efficiency while maintaining accuracy, enabling 3.98ms deep-learning-based direct wavefront reconstruction for real-time adaptive optics.