A data gluing method of quadratic polynomial fitting for lidar signal detection is proposed referring to the small gluing region and singular points. The data gluing region is extended to fully take the non-linear relationship between analog-to-digital and photon counting into account. The influence of cloud height on data quality is analyzed and the quality control is implemented to obtain stable gluing coefficient. The pile-up photon counting data is corrected according to the computational non-linear relationship between analog-to-digital and photon counting data. The synchronous comparison experiment between direct-detection Doppler lidar and balloon was carried out to verify this modified gluing method. The result shows that correlation coefficient increases from 0.80 to 0.85, the root mean square error decreased from 3.37 m/s to 2.41 m/s, and the deviation decreased from 0.60 m/s to 0.06 m/s. The result demonstrates that the modified gluing method could improve the accuracy of wind field detected by direct-detection Doppler lidar and is useful for atmosphere detection lidar to glue far-field and near-field signal.