As a passive autonomous navigation method, geomagnetic navigation has stronger stability than the Global Positioning System (GPS). To improve the accuracy of geomagnetic navigation of the UAV platform, it is needed to compensate for the fixed field interference, eddy current interference and induction interference generated by the UAV platform. Based on the measured UAV magnetic field data, the frequency characteristics of the interference signal during UAV operation are analyzed. Combining the traditional Tolles-Lawson model and Kalman filter algorithm to predict the change of geomagnetic field, and the assumption of constant geomagnetic field in the traditional Tolles-Lawson model is removed. The forgetting factor α is introduced to reduce the noise of observation covariance error and state covariance error according to the residual theory, realize noise prediction, and improve the Kalman filter. The experimental analysis shows that the improved adaptive Kalman filter effectively reduces the compensation error, and the compensation effect Improvement Ratio (IR) is 5.64 higher than that of the ordinary Kalman filter, and the magnetic field noise glitch in the compensated magnetic field data is significantly reduced.