Sows are lateral lying when they are willing to lactate, while they will turn into sternal lying or tilt lying as refusing to lactate. So the lateral lying time can be used as an important parameter to evaluate the maternal behavior of sows. Firstly, a set of monitoring nodes with sensor MPU6050 and bluetooth module was designed to monitor the behavior data at their lactation periods (28d). Then, the attitude angle in feeding time was counted, and the deflection angles of nodes were computed to achieve collision correction. Thirdly, combined with numerical features of each posture, the baseline of acceleration curve was extracted by using Haar wavelet method. Finally, the postures were divided into stand, side and tilt with the algorithm of SVM, and 10 ‘Xiaomeishan’ sows were chosen. The results showed that the deflection was reduced by the correction algorithm of node, meanwhile, the data sets were normalized and the interference of energy movement was removed by wavelet transform algorithm. The test results of 392 samples indicated that the recognition rates of lateral lying on the left and on the right were 65.8% and 90.1%, respectively, while the accuracy of lateral posture and sternal lying were 75.4% and 83.3%, respectively. Above all, the analysis of posture can effectively reflect the lactating willing of sows.