Enhancing Autonomous Orchard Navigation: A Real-Time Convolutional Neural Network-Based Obstacle Classification System for Distinguishing ‘Real’ and ‘Fake’ Obstacles in Agricultural Robotics

Agriculture, Journal Year: 2025, Volume and Issue: 15(8), P. 827 - 827

Published: April 10, 2025

Autonomous navigation in agricultural environments requires precise obstacle classification to ensure collision-free movement. This study proposes a convolutional neural network (CNN)-based model designed enhance for robots, particularly orchards. Building upon previously developed YOLOv8n-based real-time detection system, the incorporates Ghost Modules and Squeeze-and-Excitation (SE) blocks feature extraction while maintaining computational efficiency. Obstacles are categorized as “Real”—those that physically impact navigation, such tree trunks persons—and “Fake”—those do not, tall weeds branches—allowing decisions. The was trained on separate orchard campus datasets fine-tuned using Hyperband optimization evaluated an external test set assess generalization unseen obstacles. model’s robustness tested under varied lighting conditions, including low-light scenarios, real-world applicability. Computational efficiency analyzed based inference speed, memory consumption, hardware requirements. Comparative analysis against state-of-the-art models (VGG16, ResNet50, MobileNetV3, DenseNet121, EfficientNetB0, InceptionV3) confirmed proposed superior precision (p), recall (r), F1-score, complex scenarios. maintained strong across diverse environmental varying illumination Furthermore, revealed orchard-combined achieved highest speed at 2.31 FPS balance between accuracy When deployed real-time, 95.0% orchards 92.0% environments. system demonstrated false positive rate of 8.0% environment 2.0% orchard, with consistent negative both These results validate effectiveness differentiation settings. Its generalization, obstacles, make it well-suited deployment agriculture. Future work will focus enhancing improving performance occlusion, expanding dataset diversity further strengthen

Language: Английский

Shaping the Future of Horticulture: Innovative Technologies, Artificial Intelligence, and Robotic Automation Through a Bibliometric Lens

Horticulturae, Journal Year: 2025, Volume and Issue: 11(5), P. 449 - 449

Published: April 22, 2025

This study conducts a bibliometric and content analysis based on publications indexed in the Web of Science Core Collection, aiming to map evolution key themes horticultural research context technological innovation sustainability. The results reveal strong orientation toward digitalization automation, particularly through integration artificial intelligence, mechatronic systems, sensor-based monitoring crop management. In field biotechnology, keywords such as gene expression, genetic diversity, micropropagation reflect sustained interest improving resilience disease resistance vitro propagation techniques. Furthermore, concepts environmental control, soilless culture, energy efficiency, co-generation highlight focus optimizing growing conditions integrating renewable sources into protected systems. geographical distribution studies highlights increased academic output countries like India regions sub-Saharan Africa, reflecting global transferring advanced technologies vulnerable areas. Moreover, collaboration networks are dominated by leading institutions Wageningen University, which act hubs for knowledge diffusion. findings suggest that future should prioritize development durable, energy-efficient adapted various agro-climatic zones. It is recommended policymakers stakeholders support interdisciplinary initiatives, promote transfer mechanisms, ensure equitable access smallholder farmers emerging economies.

Language: Английский

CV-YOLOv10-AR-M: Foreign Object Detection in Pu-Erh Tea Based on Five-Fold Cross-Validation

Foods, Journal Year: 2025, Volume and Issue: 14(10), P. 1680 - 1680

Published: May 9, 2025

To address the problem of detecting foreign bodies in Pu-erh tea, this study proposes an intelligent detection method based on improved YOLOv10 network. By introducing MPDIoU loss function, network is optimized to effectively enhance positioning accuracy model complex background and improve small target objects. Using AssemFormer optimize structure, network’s ability perceive objects its process global information are improved. Rectangular Self-Calibrated Module, prediction bounding box optimized, further improving classification target-positioning abilities scenes. The results showed that Box, Cls, Dfl functions CV-YOLOv10-AR-M One-to-Many Head task were, respectively, 14.60%, 19.74%, 20.15% lower than those In One-to-One task, they decreased by 10.42%, 29.11%, 20.15%, respectively. Compared with original network, accuracy, recall rate, mAP were increased 5.35%, 11.72% 8.32%, improves model’s attention sizes, backgrounds, detailed information, providing effective technical support for quality control agricultural field.

Language: Английский