This study analysed the effectiveness of the YOLOv9 model for ear identification, a crucial alternative to facial recognition in biometric systems. Unlike established techniques such as the Haar cascade Viola‒Jones detector technique, and histogram of oriented gradients (HOG), the YOLOv9 model demonstrated significant improvement over conventional approaches. The model, which has a sophisticated architecture, uses pretrained weights and transfer learning to increase the detection accuracy. LabelImg was used to create a dataset of images from students aged 18–23 at the GH Raisoni College of Engineering in Nagpur, India. PyTorch was used for training on an NVIDIA GTX 1080 Ti, and data preprocessing techniques were employed to ensure diversity. YOLOv9 achieved precision of 0.95, recall of 0.90, F1-score of 0.92, and mAP of 0.93, whereas Haar Cascade’s precision of 0.88, recall of 0.82, F1-score of 0.85, and mAP of 0.86. The Viola‒Jones detector also performed well with a precision of 0.90, recall of 0.85, F1-score of 0.87, and mAP of 0.88.