Low-Altitude UAV Imagery Based Cross-Section Geological Feature Recognition via Deep Transfer Learning

Most remote-sensing systems obtain terrain surface images from vertical observation by satellites and high-altitude aircrafts. However, this can be inadequate for subsurface geological image acquisition. Indeed, subsurface or undersurface geological information is more essential for geological and civil engineering. In this work, we proposed a subsurface geological feature recognition approach: first, cross-section geological images are captured horizontally by low-altitude UAV. Later, deep transfer learning model is trained to classify the type of geological mass exposed on the cross-section. We also examined different deep transfer learning models regarding their classification performances. The result suggests that Inception-ResNet-V2 based model outperformed others by reaching an overall accuracy of 99.72%.