Abstract: In order to explore an intelligent means of detecting delamination defects in airborne warning and control system radomes made of quartz fiber reinforced polymers (QFRP), a reflective terahertz time-domain spectroscopy system based on fiber coupling was built by combining a cooperative robot. The signal noise ratio dynamic range is about 60 dB after testing. Terahertz nondestructive testing of curved structural QFRP sample with pre-buried simulated delamination defects was performed using the build system. The internal pre-buried defects were visually identified in the reflected terahertz images obtained from different scanning areas. An improved YOLOv4 algorithm was used to obtain 90.18% accuracy and 91.26% recall in automatic defect identification, which were 3.37% and 4.01% higher than the original YOLOv4 algorithm, respectively. The small target defects can also be recognized well. This experiment enriches the design and detection of airborne warning and control system radome sample, explores the detection of curved structural QFRP sample using terahertz imaging method and provides a new intelligent nondestructive testing method of airborne warning and control system radomes detection, which has the value of engineering application.