The navigation field of agricultural machinery has entered the intelligent stage, but the navigation control performance of paddy field agricultural machinery represented by rice transplanters is not stable in complex environments. Therefore, this study proposes a method to identify navigation deviation patterns based on Deep Belief Network (DBN) and designs an adaptive preview distance control method based on a driver preview model for each deviation pattern. Among them, the deviation pattern identification method is a two-stage algorithm. First, determine whether the current navigation status is abnormal. Then, the classification was refined for different abnormal states. The adaptive control method is divided into two levels. The main regulator calculates the dynamic preview distance according to the current state variable; the sub-regulator calculates the preview distance adjustment value according to the abnormal state degree. In the performance test of the identification method, all the models show excellent stability and accuracy, and the identification speed of the algorithm meets the high frequency of the rice transplanter navigation system. In the performance test of the control algorithm, compared with the static preview distance, the adaptive preview distance control method proposed in this study can effectively suppress the disturbance deviation of the rice transplanter navigation.

Aiming at the path tracking oscillation problem caused by sideslip of navigation agricultural machinery in complex farmland environments, this paper takes the Yanmar YR-70D transplanter as the research object and proposes an adaptive control method based on deep belief network (DBN) sideslip identification. Based on the preview tracking model, this method dynamically adjusts the preview distance by real-time identifying the sideslip status via DBN. In this research, the Optuna framework is utilized to optimize the DBN model, and the SHAP framework is introduced to quantify the feature contribution. The results show that, considering both real-time performance and accuracy, when the combined variable of “lateral deviation + heading deviation” with a data length of 20 is adopted, the prediction accuracy of the model reaches 88.04%, among which heading deviation (HD) has the highest contribution with a SHAP value of 0.957. Comparative experiments indicate that the accuracy of DBN is comparable to that of mainstream models such as LSTM and Transformer, but its forward propagation speed (0.0487 ms) is better than other models’. Upland field experiments verify that under different test factors of vehicle speed and slope angle, the line-on time (1.83–5.60 s), line-on distance (1.46–5.07 m), and maximum overshoot (0.32–15.75 cm) of the adaptive control group are all superior to those of the fixed-parameter control group. Paddy field experiments further confirm that under the working condition of 1.0 m/s vehicle speed and 15° slope, the above three indicators of the experimental group are 0.85–3.1 s, 0.75–3.14 m and 0–7.44 cm, respectively, all of which are better than those of the control group. The proposed method demonstrated superior path tracking performance under the operating conditions of the tested rice transplanting machines, indicating its potential application in lateral slip perception adaptive navigation control. However, further verification is still needed with different types of transplanting machines, field environments, and operating conditions to assess its general applicability.