Real-time banana freshness grading: A portable end-to-end detection system with high precision

To enhance the efficiency of banana quality management, an automated freshness grading system based on real-time image processing was developed. The system integrates a camera for image acquisition and employs portable edge devices (Nvidia Jetson Orin NX or Jetson TX2) to classify banana freshness i...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Postharvest biology and technology Jg. 231; S. 113904
Hauptverfasser: Chen, Liangyan, Zhu, Junkang, Gui, Yisong, Liu, Weihua, Zeng, Shan
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 01.01.2026
Schlagworte:
Fruit quality
Network algorithm
Non-destructive
YOLO
YOLO
AI
Network algorithm
Non-destructive
Fruit quality
ISSN:0925-5214
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:To enhance the efficiency of banana quality management, an automated freshness grading system based on real-time image processing was developed. The system integrates a camera for image acquisition and employs portable edge devices (Nvidia Jetson Orin NX or Jetson TX2) to classify banana freshness into four levels, displaying results along with confidence scores in real-time. A lightweight YOLO-based banana freshness detection method (YOLO-BFD), built on the YOLOv7-tiny framework, was designed to enable high-accuracy, real-time processing on edge devices. Key optimizations—including reduced model width factors, the LightSP module, VoVSSCSP, and Ghost convolutions—resulted in a 118.4 % increase in detection speed and a 1.4 % improvement in mean average precision (mAP) over the baseline, along with a 5 % increase in precision, 1.5 % increase in recall, and a 3 % increase in F1 score. Additionally, the system achieved a 93.5 % reduction in computational costs and a 41.9 % decrease in model parameters. The system demonstrated an average accuracy of 97.8 % in freshness grading, with real-time detection speeds of 61 frames per second on the Jetson Orin NX and 29 frames per second on the Jetson TX2. Overall, the proposed system provides a practical and efficient solution for portable banana quality assessment in agricultural applications. •The quality of bananas was assessed non-destructively using AI techniques.•A portable artificial intelligence grading and detection system was developed.•A lightweight detection network for real-detection was proposed.•A system for bananas classification after harvest is established.
ISSN:0925-5214
DOI:10.1016/j.postharvbio.2025.113904