This paper proposes an integrated UAV-SfM-AI pipeline for automated structural inspection using a Stacking Convolutional Neural Network (S-CNN) operating in texture space of 3D models. The system achieves 99.88% pixel-level accuracy in crack detection while eliminating alignment issues common in multi-view projection methods. Authors: Phromphat Thansirichaisree, Apichat Buatik (Thammasat University, Thailand). Presented at TSER2025 Symposium, Bangkok, August 2025.