A scene-adaptive reseeding system with missed seeding detection for double-disc air-suction seed meter based on an improved YOLOv8n algorithm

In precision seeding operations, the detection of missed seeds and the subsequent reseeding process are pivotal for enhancing seeding accuracy and maximizing crop yield. This study presents a scene-adaptive reseeding system with missed seeding detection for a double-disc air-suction seed meter, based on an improved YOLOv8n algorithm. The system is composed of several integral components, including a missed seeding detection module, the double-disc air-suction seed meter, a scene-adaptive reseeding module, and a central controller, among others. Industrial cameras are employed to dynamically capture images within the seed meter, which, in combination with the refined YOLOv8n model, ensures precise detection of missed seeding events. The system identifies 18 distinct missed seeding scenarios and incorporates six distinct dual-phase reseeding strategies tailored to address varying operational conditions. The central controller, guided by detection results and preset logic, actuates a permanent magnet synchronous motor (PMSM) through field-oriented control (FOC), enabling adaptive reseeding and ensuring stable motor operation with a millisecond-level response. Ablation tests reveal that the enhanced YOLOv8n model improved detection precision (P) by 4 % over the original model, while its inference speed surged to 303.03 FPS. This enhancement significantly bolstered both detection accuracy and speed, while maintaining the model’s lightweight nature. Bench tests indicated that, within a speed range of 8–16 km/h, the system achieved an average monitoring accuracy exceeding 91 % for seeding count, reseeding count, and missed seeding count. The minimum detection accuracy for damaged seeds was 82.86 %. Upon incorporating the compensation system, seeding accuracy increased by 0.15 % to 2.89 %. The most significant compensation effect was observed within the 10–14 km/h speed range, where the maximum increase approached 3 %. Kruskal–Wallis test results indicated that speed had a marked influence on reseeding performance; at lower speeds, overcompensation occurred, while higher speeds led to a decrease in reseeding accuracy. The missed seeding detection and scene-adaptive reseeding system proposed herein proved both effective and feasible, demonstrating superior accuracy and stability when operating within a speed range of 10–14 km/h. It meets the requirements for missed seeding detection and adaptive reseeding for double-disc air-suction seed meters, thus contributing significantly to the advancement of automation and intelligent systems in high-speed precision seeding operations.