Additive Manufacturing (AM) has become essential in renewable energy, medical devices, and aerospace. Ensuring defect-free AM components is critical for durability and safety, making defect detection key for quality assessment. However, accurate real-time defect detection remains challenging due to technology limitations, particularly in detecting multiscale defects. Recent deep learning (DL) advancements have transformed AM defect detection by enabling end-to-end solutions that simplify traditional processes. We proposed a DL-enhanced defect detection method optimized for real-time detection during AM processes. A DL model was developed and trained on a comprehensive AM defects dataset, incorporating data augmentation to enhance model robustness and address class imbalance. This model was designed to handle textures and irregularities characteristic of AM. Our approach identifies various defect types, including scratches, cracks, warping, and material inconsistencies, addressing multiscale defect detection challenges in AM procedures. Experimental results demonstrated that this method achieved a mean average precision () of 0.438 with high performance for defect classes Curling_Spaghetti (74.3% mAP) and Warping (55.9%). This performance proved the method's capability for reliable automated quality assessment in AM.