Dance is constantly discovering truth, goodness, and beauty in human social life, spreading truth, goodness, and beauty, and fully expressing the artistic pursuit of dance beauty. It shapes different dance images, expresses the aesthetic consciousness and feelings of dance, and resonates with the audience to meet their aesthetic needs through various forms of movement. Because the RBF neural network model is good at approximating functions, many researchers have begun to use the RBNN approximation model for engineering design. Due to the limited dance data available for research, this paper uses radial basis function neural network model to study the aesthetic characteristics of dance in the context of few-shot learning. When the time index reaches 50, the average ratio of the L-MBP algorithm is 33.4 percent, 32.5 percent for the RBNN algorithm, and 46.3 percent for this method. As can be seen, this method has the highest ratio of the three algorithms, giving it a distinct advantage in terms of dance aesthetics. As a result, this paper establishes a neural network model, trains and simulates the network model, studies and analyzes the influence of changes in influencing factors on the aesthetic characteristics of dance, and provides a new idea for the prediction of the aesthetic characteristics of dance and a reference for optimizing the design of the aesthetic system of dance using the prediction ability of radial basis function neural networks.