The connectivity of natural fractures is conducive to reservoir reconstruction and can be used as a connecting channel for different oil and gas reservoir units, which is the research focus of exploration and development. The existing fracture connectivity calculation is mainly applied to the connectivity analysis of regional fault system, which relies on the quality of seismic data and imaging logging. The data requirements are high, and the transformation of micro-fractures to reservoirs is neglected. In order to solve the above problems, a slice connectivity calculation method based on semantic segmentation deep learning was proposed. Taking the thin section image of the tight sandstone reservoir of Huangliu Formation in Qiongdongnan Basin as an example, the connectivity of thin section fractures was calculated. Compared with the traditional manual identification, this method has higher efficiency. In addition to the inaccurate identification of some fractures, the overall calculation error was less than 10%, which has a good application prospect. The fracture connectivity of tight sandstone reservoirs in different fault sections of Huangliu Formation in Qiongdongnan Basin was calculated, and the connectivity of Songtao section > Lingshui section > Yaxi-Yanan section was obtained, which was in good agreement with the degree of fault development. This method combines artificial intelligence technology and geological information to effectively calculate the connectivity of micro-fractures at the slice scale, which is helpful to clarify the transformation effect of micro-fractures on reservoirs. Moreover, this method has low data requirements and is suitable for the evaluation of reservoir properties in the early stage of exploration and development.