In this paper, for the first time, a one-dimensional convolutional neural network using a U-shaped architecture is evaluated in the context of radar cross section (RCS) based chipless RFID (CRFID) systems. A 3-bit CRFID tag is utilised to create eight discernible RCS signatures representing identification numbers. A dataset of 9,600 measured RCS signatures was utilised for training, validating, and testing the model. The dataset was collected by placing the tag on varying surface shapes, orientations, and read ranges to enable robust detection. The root mean square error (RMSE) metric was used to assess the model’s performance. The achieved RMSE was 0.11 (1.5%). The low RMSE score demonstrates the effectiveness that this type of architecture has in accurately detecting and generalizing the encoded information from the RCS signatures.