Abstract: To address the challenges of poor accessibility in contact detection and difficulty in water immersion scanning path planning for defects at the corners of carbon fiber reinforced polymer (CFRP) truss joints, a guided wave detection method was proposed. The method employed a pitch-catch configuration to excite and receive guided waves. Wavelet packet decomposition was applied to the guided wave signals to generate feature maps, and a neural network was utilized for defect identification. A dedicated detection system for CFRP truss joint defects was developed. Experiments were conducted on CFRP specimens with prefabricated flat-bottom holes and delamination defects at the corners. A dataset was constructed to train a convolutional neural network model, and validation tests were conducted on the detection system. Results demonstrated detection accuracies of above 99% for delamination defects of 6 mm or larger and flat-bottom hole defects of 2 mm or larger, validating the effectiveness of the method in addressing defect detection challenges for CFRP truss joints.