Studying DFS river parameters has become crucial for predicting oil and gas reservoirs and guaranteeing geological work since the concept of distributive fluvial systems （DFS） was introduced. However，manual measurement methods are time-consuming and labor-intensive. To automatically extract，study，and characterize DFS river parameters，this paper proposed a distributive fluvial system river identification and parameter extraction method based on deep learning and morphology. First，a network model，Seg_ASPP，integrating Segformer and ASPP，was proposed for generating river network masks. Subsequently，the cumulative cost and polynomial fitting algorithms were used to extract the river mask centerline，and a scheme was designed to extract the DFS river length，width，and curvature based on the mask centerline. Finally，the DFS area in the Golmud region of the Qaidam Basin was selected as the experimental area，and multiple deep learning models were used to predict DFS rivers. The accuracy of the masks extracted by the Seg_ASPP network model was compared and evaluated. The automatically extracted parameter values obtained using this method were compared with manually measured actual values，showing average relative errors of 10.22%，13.57%，and 5.41% for length，width，and curvature，respectively. The DFS river fan in the Golmud experimental area was quantitatively characterized by the grid method. According to the different changes in the distance of the parameters from the vertex，the DFS was divided into braided river sections，braided and meandering sections，and low-bend river sections.