TY  - JOUR
AU  - Ma, Zhenhao
AU  - Zhang, Bin
AU  - Yin, Tianzhen
PY  - 2026
DA  - 2026/04/17
TI  - A Robotic System for Fine-Grained Non-Destructive Grading of Visually Similar Fruits Based on Improved YOLOv11 and Multi-modal Perception
JO  - ICCK Transactions on Intelligent Systematics
T2  - ICCK Transactions on Intelligent Systematics
JF  - ICCK Transactions on Intelligent Systematics
VL  - 3
IS  - 2
SP  - 94
EP  - 107
DO  - 10.62762/TIS.2025.566749
UR  - https://www.icck.org/article/abs/TIS.2025.566749
KW  - fruit grading robot
KW  - YOLOv11
KW  - fine-grained classification
KW  - NIR spectroscopy
KW  - flexible grasping
AB  - To address key challenges in post-harvest fruit grading—namely the difficulty of distinguishing visually similar varieties, the invisibility of internal quality, and mechanical damage during grasping—this study develops an intelligent robotic grading system that integrates advanced computer vision, Near-Infrared (NIR) spectroscopy, and flexible force-controlled grasping. First, an improved object detection algorithm, YOLOv11-TFE, is proposed to mitigate visual confusion between Qixia Fuji apples and Beijing Pinggu peaches and to handle the irregular geometry of Nanshui pears. By embedding the parameter-free SimAM attention mechanism into the backbone to explicitly enhance and decouple surface texture features (gloss versus tomentum) and incorporating the DySample upsampling operator to better preserve complex edge information, the discriminative capability of the detector is significantly improved. Second, a non-destructive internal quality detection module based on NIR spectroscopy (650–1100 nm) is constructed. Through the optimization of preprocessing strategies—specifically Improved Derivative Correction (IDC) and Standard Normal Variate (SNV)—robust PLSR models for soluble solids content (SSC) prediction are established for fruits with differing skin textures. Finally, a dynamic actuation unit for a biomimetic flexible gripper has been designed to achieve non-destructive sorting under continuous flow conditions. Experimental results show that the improved visual algorithm achieves an average [email protected] of 94.6%, with detection accuracies for apples and peaches reaching 96.0% and 97.8%, respectively, markedly reducing inter-class confusion. The root mean square error of prediction (RMSEP) for SSC across all three fruit varieties is kept within 0.65%. System-level validation further demonstrates an overall dynamic grasping success rate of 91.7% without causing visible damage. Overall, the proposed system achieves precise and comprehensive grading for multiple high-value fruit varieties.
SN  - 3068-5079
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Ma2026A,
  author = {Zhenhao Ma and Bin Zhang and Tianzhen Yin},
  title = {A Robotic System for Fine-Grained Non-Destructive Grading of Visually Similar Fruits Based on Improved YOLOv11 and Multi-modal Perception},
  journal = {ICCK Transactions on Intelligent Systematics},
  year = {2026},
  volume = {3},
  number = {2},
  pages = {94-107},
  doi = {10.62762/TIS.2025.566749},
  url = {https://www.icck.org/article/abs/TIS.2025.566749},
  abstract = {To address key challenges in post-harvest fruit grading—namely the difficulty of distinguishing visually similar varieties, the invisibility of internal quality, and mechanical damage during grasping—this study develops an intelligent robotic grading system that integrates advanced computer vision, Near-Infrared (NIR) spectroscopy, and flexible force-controlled grasping. First, an improved object detection algorithm, YOLOv11-TFE, is proposed to mitigate visual confusion between Qixia Fuji apples and Beijing Pinggu peaches and to handle the irregular geometry of Nanshui pears. By embedding the parameter-free SimAM attention mechanism into the backbone to explicitly enhance and decouple surface texture features (gloss versus tomentum) and incorporating the DySample upsampling operator to better preserve complex edge information, the discriminative capability of the detector is significantly improved. Second, a non-destructive internal quality detection module based on NIR spectroscopy (650–1100 nm) is constructed. Through the optimization of preprocessing strategies—specifically Improved Derivative Correction (IDC) and Standard Normal Variate (SNV)—robust PLSR models for soluble solids content (SSC) prediction are established for fruits with differing skin textures. Finally, a dynamic actuation unit for a biomimetic flexible gripper has been designed to achieve non-destructive sorting under continuous flow conditions. Experimental results show that the improved visual algorithm achieves an average [email protected] of 94.6\%, with detection accuracies for apples and peaches reaching 96.0\% and 97.8\%, respectively, markedly reducing inter-class confusion. The root mean square error of prediction (RMSEP) for SSC across all three fruit varieties is kept within 0.65\%. System-level validation further demonstrates an overall dynamic grasping success rate of 91.7\% without causing visible damage. Overall, the proposed system achieves precise and comprehensive grading for multiple high-value fruit varieties.},
  keywords = {fruit grading robot, YOLOv11, fine-grained classification, NIR spectroscopy, flexible grasping},
  issn = {3068-5079},
  publisher = {Institute of Central Computation and Knowledge}
}