Abstract: The quality of seeing is a decisive factor affecting the image quality of astronomical observations. Currently, daytime seeing data is mainly obtained through the Solar Differential Image Motion Monitor (SDIMM) or spectral ratio method. Due to the non-identity between the SDIMM and the actual observation instrument, it cannot reflect the actual seeing at the time of data acquisition, nor trace the seeing corresponding to historical existing observation data. The spectral ratio method requires a large amount of short-exposure data, the computational cost is huge. Based on the above difficulties faced by astronomical observations, this paper proposes a neural network-based daytime seeing prediction method. This method first uses the spectral ratio method to calculate the corresponding seeing r_0 for the obtained short-exposure data, and constructs the data. Then, the principal component analysis method was used to reduce the dimensionality of the data, and the nonlinear regression relationship between the observation image of the solar photosphere with the narrow-band filter and the seeing degree was established through the neural network. The experimental results of the training set and the test set show that this method can be used to estimate the seeing. Using this method to estimate the seeing of HSOS (Huairou Solar Observing Station) in 2020, the median visual acuity was 2.89 cm. This method was used to perform the long-term statistical analysis of long-term seeing of historical observation data from 1989 to 2010 for 22 consecutive years, and the results showed that the median visual acuity of HSOS was around 3 cm. Seeing above 3 cm exceeded 40%. The best seeing was observed in September of the year, which confirmed the long-term stability of visual acuity at HSOS. In addition, this method can also provide a judgment basis for selecting high-quality short exposure image frames based on rapid judgment of seeing r_0.