RNA-binding proteins (RBPs) are pivotal in regulating various biological processes, including RNA splicing and protein translation. Accurate classification of RBPs is crucial for understanding their functional roles in cellular mechanisms. Here we present PIONet, a deep learning method based on a convolutional neural network (CNN) that accurately classifies RBPs. Traditional one-hot encoding methods fail to capture positional relationships within RNA sequences. The PIONet addresses this limitation and enriches sequence representations with spatial information by integrating positional encoding with one-hot encoding. The CNN model processes these combined features to extract local patterns and motifs critical for RNA-protein interactions. The positional encoding enables the model to effectively learn inter-nucleotide dependencies within the sequence, making it highly suitable for capturing the intricate patterns associated with RBPs. Our model has been evaluated on 24 benchmark datasets, collectively referred to as RBP-24, demonstrating significant improvements in classification accuracy and robustness compared to the state-of-the-art methods. Additionally, we applied transfer learning to enhance the model’s performance, which led to improvements on 23 out of 24 datasets. This framework presents a powerful tool for advancing RBP classification tasks and offers potential applications in understanding post-transcriptional gene regulation.