RNA has the potential to serve as a drug target, requiring RNA-small molecule binding affinity to screen potential drugs generally. However, accurately predicting RNA-small molecule binding affinity remains a highly challenging task. This study proposes an explainable multiview, multiscale deep learning network, EMMPTNet, to address these challenges based on physicochemical and topological properties. EMMPTNet efficiently extracts features from multiple views through four modules, and a multilayer perceptron is employed to predict binding affinity based on the multiview, multiscale features extracted by these modules. Experimental results show that EMMPTNet outperforms current methods with a mean absolute error (MAE) of 0.058 and a Pearson correlation coefficient (PCC) of 0.773. To demonstrate the model's interpretability, this study provides an analysis of the feature extraction process across different views and visualizes the overall feature importance distribution combining all views. Furthermore, validation studies on newly discovered RNA-small molecule compounds further confirm the generalization ability of EMMPTNet.