Fruit-YOLO11: A Lightweight Framework for Fine-Grained Detection in Assessing Fruit Freshness

Read full text Summarise with VIRAT
Authors: Zihong Wu, Yuexing Wu
Cited by: 0
References: 29

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

In intelligent sorting scenarios within agricultural supply chains, the early mild spoilage of fruits often manifests solely as high-frequency micro-textural changes, including mycelia, microcracks, and slight browning.These changes are easily masked by extensive areas of healthy peel.Simultaneously, detection models deployed in resource-constrained environments generally experience a coupled contradiction: the coexistence of spatial computational redundancy with the loss of fine-grained features.To achieve efficient and accurate detection for the fine-grained assessment of fruit freshness on edge devices, a lightweight framework for detection, called Fruit-YOLO11, was proposed in this study.By integrating adaptive downsampling and partial convolution, the backbone and neck networks of YOLO11 were structurally reconstructed.This reconstruction established a mechanism for spatial redundancy suppression and efficient feature transmission that specifically target healthy peel regions.Furthermore, an attention-driven intrinsic feature integration head was designed to dynamically recalibrate and enhance microscopic pathological features, including early mycelia, physiological microcracks, and slight browning, which are prone to attenuation during feature compression.The effectiveness of each module was verified through comparative and ablation experiments on a multi-class dataset for fruits.Results demonstrate that, when the number of parameters and computational complexity are reduced by 21.6% and 25.4%, respectively, compared with the baseline model, Fruit-YOLO11 maintains a real-time inference speed of 70.9 frames per second, and detection accuracy increases to 78.6%.The proposed framework exhibits higher recognition sensitivity to the fine-grained features of early mild spoilage and a lower risk of missed detections.This study provides a technical solution that balances real-time performance and accuracy for the large-scale assessment of fruit freshness.This solution can be employed in low-cost edge computing devices within automated sorting lines for smart agriculture.

Fields of Study & Subjects

Smart Agriculture and AI Advanced Neural Network Applications Industrial Vision Systems and Defect Detection

Publication details

Journal / source
Journal of Engineering Science and Technology Review
Volume
19
Issue
1
Pages
226–236
Publication date
1 February 2026
Language
EN