Related Experiment Video
Updated: Apr 9, 2026
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.7K
Lightweight Multi-Occlusion Pear Detection via Multi-Auxiliary Domain Transfer Learning for Robotic Harvesting.
Pengfei Lv1, Jinlin Xue1, Shaohua Liu1
1College of Engineering, Nanjing Agricultural University, Nanjing, China.
Annals of the New York Academy of Sciences
|April 7, 2026
Summary
A new lightweight detector improves occluded pear detection for robotic harvesting. Multi-auxiliary domain transfer learning (MADTL) enhances accuracy and speeds up training, enabling efficient deployment on edge devices.
Area of Science:
- Agricultural robotics
- Computer vision
- Machine learning
Background:
- Accurate detection of occluded pears is crucial for efficient robotic harvesting.
- Existing methods struggle with lightweight design, domain shift in transfer learning, and multi-category classification accuracy.
Purpose of the Study:
- To develop a lightweight, accurate, and efficient detector for multi-category occluded pear detection.
- To address challenges in model size, training time, and classification accuracy for robotic picking applications.
Main Methods:
- Proposed a lightweight detector built upon YOLOv8, optimizing backbone and neck architectures.
- Implemented multi-auxiliary domain transfer learning (MADTL) using apple and orange datasets to bridge domain gaps.
- Integrated advanced modules for enhanced feature extraction and fusion efficiency.
Main Results:
- Reduced model size by 62.4% and floating-point operations by 53.7% compared to YOLOv8s.
- MADTL accelerated convergence by 75% and improved detection accuracy.
- Achieved real-time inference at 47.39 ms per image in field deployment.
Conclusions:
- The proposed lightweight detector with MADTL enables efficient, accurate, and real-time detection of occluded pears.
- This technology supports selective harvesting on resource-constrained edge devices, minimizing picking failures and enhancing operational efficiency.
