Precise geolocation algorithm for multi-target plants in banana plantations based on synchronized GCPs and image point estimation

【Objective】This study aimed to establish a precise geolocation algorithm for multi-target plants in banana plantations based on synchronized ground control points （GCPs） and image point estimation （SGIE-MPL）， thereby providing technical support for tasks such as single-plant-level precision management in banana cultivation， autonomous path planning and navigation of agricultural machinery and robots.【Method】The SGIE-MPL algorithm was established based on synchronized GCPs and image point estimation to achieve concurrent localization of targets with introduction of synchronized GCPs， and camera field of view modeling was utilized to eliminate cumbersome calibration procedures. Pixel coordinates of banana plant base points were estimated to address plant height-induced problems （non-coplanarity between target points to be localized and GCPs and localization accuracy affected by projection error） for single-plant-level precise geolocation. Based on this， the camera and terrain model-based localization algorithm （CTLA） and the monocular vision-based object localization model （MV-OLM） were compared， the localization accuracy of SGIE-MPL algorithm was tested under multi-factor coupled conditions， and the stability was tested by performing multi-target engineering trials.【Result】At unmanned aerial vehicle （UAV） flight altitudes of 20 m and 30 m， the SGIE-MPL algorithm achieved localization accuracy with an excess indicator value of 0.04 m， far lower than the mean leaf length of banana； the mean absolute error （MAE） and root mean square error （RMSE） were less than 3.00 m across test groups， with the algorithm compliance rate （P） of 98.52%. Under the same test conditions， the SGIE-MPL algorithm outperformed CTLA and MV-OLM algorithms with similar technological pathways in terms of three indicators （MAE， RMSE， and P）. Under multi-factor coupled conditions such as plant height and UAV flight altitude （20-50 m）， the SGIE-MPL algorithm demonstrated superior robustness and localization accuracy. According to the results of SGIE-MPL multi-target engineering test based on large-scale banana plantation， for the 462 target points to be localized， the target detection image aligned with the map， indicating that the SGIE-MPL algorithm could calculate all geographic coordinates of targets stably.【Conclusion】The SGIE-MPL integrated software platform based on synchronized GCPs and image point estimation can realize single-plant-level precise geolocation in the dense multi-target scene of banana plantation， with the P of 98.52%， which ob-viously outperforms the CTLA and MV-OLM algorithms. It requires only consumer-grade UAVs to provide technical support for tasks such as single-plant-level precision management in banana plantations， autonomous path planning and navigation of agricultural machinery and robots.