Smart vision-based sugarcane bud detection and cutting system for seed generation

• Developed an autonomous, low-cost sugarcane bud detection and cutting system (VSBD-CU) using YOLOv8-nano for real-time image-based detection and a Raspberry Pi 4 for onboard control and actuation. • Achieved high detection accuracy ([email protected] = 98.1 %) and fast detection speed (45 fps) even under variable lighting and without pre-cleaned or manually marked stalks. • Optimized mechanical cutting unit with interchangeable blade bevel angles, with 30° bevel providing the best results across sugarcane varieties, minimizing bud injury (<2.5 %) and ensuring clean separation. • Cutting throughput of 3600 buds/hour, with average detection time of 0.90 s for mature sugarcane. • Two-way ANOVA revealed blade bevel angle and variety significantly affect cutting time and efficiency, with strong interaction effects ( p < 0.001), justifying the system’s modular blade design. • Validated system performance across three commonly cultivated Indian sugarcane varieties (CoX 20,055, CoP 18,437, CoX 20,246) without the need for stalk stripping or artificial markers. Sugarcane is an economically significant agricultural crop grown in tropical and subtropical areas for the production of sugar, ethanol, biofuel, and related by-products. The extraction of viable buds for planting is crucial to its propagation. Conventional bud cutting requires considerable manual effort, is labor-intensive and is prone to inconsistencies. This research presents a Vision-Based Sugarcane Bud Detection and Cutting Unit (VSBD-CU) designed to automate the processes of bud detection and cutting through the integration of computer vision and mechatronics. The system employs YOLOv8, for bud localization, along with a Raspberry Pi-4B managed electromechanical cutting mechanism. The system demonstrates high performance with mature buds, achieving a mean Average Precision ([email protected]) of 98.1 %. Real-time detection ranges from 0.87 to 0.95 s, with a cutting throughput of up to 1696 buds per hour. Detection efficiency for immature buds was observed to be lower, averaging 25.56 %, attributed to obstructions such as the leaf sheath. Comprehensive experimental assessments were performed to examine the influence of blade bevel angles (30°, 45°, and 60°) and sugarcane varieties (CoX 20055, CoP 18437, and CoX 20246) on cutting rate and duration. ANOVA statistical analysis indicated that blade bevel angle has a significant effect on cutting rate ( F = 2268.49) and cutting time ( F = 5555.71), with significant interaction effects observed between blade angle and cane variety ( p < 0.001). A blade angle of 30° consistently provided optimal performance across all varieties. The proposed system demonstrates significant potential for improving precision agriculture via future robotic integration and field implementation.