Abstract: Long-distance oil and gas pipelines passing through the areas with serious geological hazards are prone to bending deformation due to the external soil loads, which will threaten the safe operation of pipelines. The inline inspection technology based on the inertial measurement unit (IMU) is the main means to inspect the local deformation of pipelines at present. Herein, the characteristics of IMU data of the four typical locally-deformed pipeline sections, i.e. the buried pipeline elbows, the dented pipelines, the pipelines with bending deformation and the abnormal girth welds, were provided. Meanwhile, a IMU data pre-processing method based on wavelet denoising was put forward, a deep neural network model was established to identify the IMU data thermal map of the 4 types of typical locally-deformed pipeline sections, and a set of method was developed to identify the pipeline sections with bending deformation based on IMU data. By analyzing the 6-year IMU data of China-Russia Crude Oil Pipeline with the new method, totally 33 177 data of sample pipeline sections were formed, and an IMU bending strain database was set in China. The results of application example show that: the accuracy of the deep neural network model established based on the database to identify the pipeline sections with bending deformation is more than 90%, and the identification efficiency is up to 0.02 min/km. Hence, the method for identification of pipeline sections with bending deformation based on IMU data provides an effective technical means to identify the deformed pipeline sections with bending strain greater than 0.125% in the integrity assessment of pipelines.