U-Net Enhanced Depth Sensing Detection on Earthwork Management

Efficiently managing the substantial annual earthwork volume of more than 30 million m3 in Taiwan demands considerable resources and human effort. Practitioners in this field seek a streamlined solution for effective earthwork management. This study aims to enhance earthwork management through computational intelligence by employing U-Net for depth sensing detection. The literature review encompasses earthwork calculation, the feasibility of unmanned aerial vehicle (UAV) detection, object detection principles, and U-Net configuration. Employing a UAV, aerial depth sensing is executed at a predetermined earthwork site spanning 43,939 m2. The UAV maintains a steady altitude of 60 m above sea level, traveling at a speed of 4.7 m/s, resulting in a total airborne time of 10 min and 51 s. Capturing 276 high-resolution images with 80% overlap and 70% photo overlap, the U-Net model achieves an average detection accuracy of 92.49% for earthwork site identification. A comparison with actual earth imported to the site over 10 months reveals a 13.84% increment in detected volume by the proposed model. These findings, aligned with empirical operations, offer practitioners an efficient approach to earthwork calculation and management.