In the field operation of agricultural machinery automatic navigation system, the windbreak trees of field edge will have strong disturbance to the satellite signal. Because of the requirement of accuracy for agricultural machinery navigation and the general automatic navigation system has poor resistance to environmental interference, the algorithm of heading information fusion was studied based on the integrated navigation system of INS/GNSS. This algorithm adopted adaptive Kalman filter to reduced the noise with the measurement of heading data for single antenna GNSS and obtained the error estimation of heading data by using compensated Kalman filter. According to the quality of GNSS signal and the heading angle gradient, it could also reasonably allocate the weights of different fusion data by the calculation of federated filter. The results of simulation experiment and actual application test showed that: taking the heading measured value of double antenna GNSS as the reference data, the average absolute error of fusion heading data was -0.02° and the standard deviation was 0.50° in the process of linear driving. In the process of steering driving, the average absolute error of fusion heading data was 0.62° and the standard deviation was 2.42°. The accuracy of heading output after fusion had an obvious improvement when compared with using INS or GNSS separately. The noise of heading measured value of GNSS was filtered from the output of fusion heading and the update rate of GNSS calculated value was increased at the same time. The algorithm of heading information fusion could enhance the accuracy of the automatic navigation system for agricultural machinery and give full play to the advantages of the INS/GNSS integrated navigation system.