CycleGAN-Based Data Augmentation for Enhanced Radar-Based Tree Defect Detection

Ground penetrating radar has become a prominent tool for non-destructive health assessment of trees. Particularly, it has been used for imaging the tree roots to distinguish the healthy and rotten roots and the interiors of the tree trunks to detect structural defects (e.g., cavity and decay). Recently, a stand-off radar has been proposed to rapidly detect tree defects inside tree trunks (J. Qian, Y. H. Lee, K. Cheng, Q. Dai, M. L. M. Yusof, D. Lee, and A. C. Yucel, “Deep Learning-Augmented Stand-off Radar Scheme for Rapidly Detecting Tree Defects,” IEEE Transactions on Geoscience and Remote Sensing, vol. 62, Jun. 2024). This radar is operated on a straight slider for very rapid scanning and leverages novel deep learning techniques for real-time detection of the defects inside tree trunks, while achieving over 90% and 75% detection accuracy for tree trunk samples and living trees, respectively. However, the detection accuracy of such radar on living trees is limited due to the small size of the training dataset, since collecting radar measurements is time-consuming and labour-intensive. To this end, an effective data augmentation technique for generating a large dataset is needed, which will also be beneficial for large dataset generation for other AI-based radar detection and imaging technologies.