Lane detection is a key technology in intelligent driving. It is of great significance to detect lane position quickly and accurately to improve the safety of driving vehicles. Therefore, an improved lane detection method based on row direction position classification is proposed. The coordinate attention mechanism is integrated into the feature extraction backbone network to enhance the weight of effective positions in the feature map. Secondly, ELAN module and MP subsampling module are introduced to improve the feature extraction capability of the model. In reasoning, the idea of structure re-parameterization is used to fuse convolution and BN layer to speed up reasoning. In order to verify the performance of the improved model, the improved model was tested on TuSimple and CULane two classical lane data sets, and the detection accuracy was increased by 0.09% and 2.5% respectively compared with the original model, which verified the effectiveness of the improved model.