A neural network-based system designed for automated detection of concealed items using a postal terahertz scanner is presented, with system optimization provided. A dataset of objects scanned by a THz scanner is introduced. A convolutional neural network is trained on this dataset of terahertz images to classify and detect whether an image contains a prohibited item or not. The system is tested using real-world samples, achieving an accuracy of 95.5% mAP@0.5. The results demonstrate the effectiveness of employing a neural network in postal terahertz scanners and its potential for use in security and surveillance applications.