A detection system based on multi-sensor data fusion for corn seeding depth detection in a closed trench

• Develop a multi-sensor based detection system to innovatively detect the seeding depth of seeds in a closed trench. • A seeding depth prediction model based on BP neural network optimized by PSO and GWO is constructed. • Select parameters that represent the vertical movement behavior of no-till precision planter as the predictors of the model. • The model has a relatively high prediction accuracy for seeding depth. Seeding depth is a critical indicator for evaluating the quality of no-till precision corn seeding, and significantly influences the corn emergence time and final yield. However, accurate measurement of the seeding depth in a closed trench during the seeding process remains a challenge in the field of precise corn seeding. In this study, a seeding depth detection system based on multi-sensor data fusion was developed to measure seeding depth in a closed trench. A seeding depth prediction model was established based on the working speed, the swing angle of the parallel four-bar linkage, the three-axis angles and acceleration of the gauge arm, and the downforce exerted by the gauge wheel on the ground as inputs, and the seeding depth as the output. The modelled dataset was derived from indoor soil bin experiments. Three algorithms, moving average filtering, Kalman filtering and Savitzky-Golay (SG) filtering, were used for noise reduction on the force sensor data. The results show that SG filtering has the best noise reduction performance. Comparison of the predictive performance of the GWO-BP and PSO-GWO-BP models showed that the PSO-GWO-BP model had the best modelling performance with R 2 = 0.92, MAE = 0.03 and RMSE = 0.08. Further evaluation of the developed prediction model through field experiments showed that the PSO-GWO-BP model could accurately predict the seeding depth in a closed trench with R 2 = 0.91, MAE = 0.03 and RMSE = 0.08. The progress demonstrated in this study is of great significance in promoting the informatization of seeding quality and the intelligence of seeding operations.