Smart solutions for capsicum Harvesting: Unleashing the power of YOLO for Detection, Segmentation, growth stage Classification, Counting, and real-time mobile identification

Abstract

This research paper explores a comprehensive approach to advancing capsicum harvesting by integrating cutting-edge technologies. The study addresses key objectives, including capsicum detection using various YOLO algorithms, peduncle detection through YOLO segmentation models in a proposed robotic harvester, laboratory testing of cutting target point coordinates using the Real Sense D455 RGB-D camera, growth stage determination, and capsicum counting/tracking with a supervision algorithm. The investigation highlights the YOLOv8s model as the most successful for capsicum detection, achieving a remarkable mean Average Precision (mAP) of 0.967 at a 0.5 Intersection over Union (IOU) threshold. As part of the growth stage determination task, YOLOv8s achieved a satisfactory mAP of 0.614 at the same IOU threshold. Additionally, the YOLOv8s-seg model demonstrated superior performance in peduncle detection, attaining a box mAP of 0.790 and a mask mAP of 0.771. The YOLOv8s-seg model excels in peduncle detection with a box mAP of 0.790 and a mask mAP of 0.771. Laboratory experiments using the Real Sense D455 RGB-D camera showcased its capability to localize the target point with a maximum error of 8 mm longitudinally, 9 mm vertically, and 12 mm laterally. The developed tracking and counting algorithm achieve a notable counting accuracy of 94.1 % during the third harvesting cycle in the greenhouse. The Android application developed demonstrated robust performance, achieving high accuracy (Mean IoU: 0.92), precise localization (Mean Euclidean Distance: 5 pixels), responsive user interface (Touch Response Time: 150 ms), and broad compatibility across Android versions and device types, with effective error handling (Success Rate: 95 %). The study not only advances capsicum harvesting techniques but also presents practical insights for the integration of advanced technologies, paving the way for efficient robotic harvesting systems in agriculture.

Plant phenotyping relevance

YOLO画像解析、成長段階分類、果実カウント・追跡を統合した植物状態・器官形質の取得手法を開発し、精度を評価しているため、収穫ロボットの単なる対象定位を超える中心的なフェノタイピング研究である。

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.