Vision-mechanical collaborative continuous hole fertilization system for maize root-zone: Side-position dynamic variable fertilization and real-time positioning of root-zone

Uneven plant spacing in maize fields reduces the reliability of conventional hole-fertilization methods that depend on preset plant spacing, mechanical synchronization, or single-point triggering for precise fertilizer placement, leading to deviations in fertilization timing and root-zone positioning. In this study, a vision-mechanical collaborative system for continuous variable hole fertilization in maize root zones was developed to enable adaptive fertilization based on real-time root-zone position. The system integrates a dynamic variable hole-fertilization actuator, the RootLocNet root-zone perception and localization module, and the FTC collaborative control model for continuous hole fertilization. The actuator uses a dual-station external-groove-wheel fertilizer metering rotor and a fertilizer discharge baffle to achieve alternating variable fertilizer discharge within 3.0–6.0 g. RootLocNet is based on YOLOv11n and incorporates the MSCA attention mechanism. Combined with DeepSORT tracking and stereo-depth conversion, it enables three-dimensional root-zone localization and target ID maintenance. The FTC model converts root-zone localization results and mechanism response time into fertilizer-discharge trigger timing, thereby enabling side-position variable hole fertilization of the root zone under continuous travel. The experimental results showed that RootLocNet achieved an mAP50, mAP50–95, and precision of 95.5%, 43.5%, and 93.3%, respectively, with a single-frame inference time of 4.15 ms. The average R² between the root-zone localization results and measured values was 0.9553, with an RMSE of 17.9860 mm. Field trials showed that, at operating speeds of 0.4–1.0 m/s, the fertilizer placement error was 33.6–46.1 mm, the fertilization length was 76.4–128.8 mm, and the coefficient of variation of fertilization-length stability was 4.44%-7.62%. These results indicate that vision-mechanical closed-loop coordination based on real-time three-dimensional root-zone localization can reduce the dependence of continuous hole fertilization on preset plant spacing and improve the positional adaptability and execution stability of root-zone fertilization under non-uniform maize spacing.