Scholarlink

<p>The precise detection and classification of pineapples at various stages of maturity are crucial for early yield estimation in agriculture. This study presents a framework utilizing YOLOv8 for pineapple detection and ripeness classification through drone imagery. After detecting and classifying pineapples, we employ bounding box and binary mask fitting techniques, including Circle Fitting (CF), Ellipse Fitting (EF), Circle Enclosing (CE), and Rotated Rectangle (RR), to estimate pineapple size by converting pixel-based measurements to millimeters. Finally, the current research predicted the weight of the pineapples by employing linear and nonlinear models. This work used linear models such as Linear Regression, Polynomial Regression, and nonlinear-based machine learning models such as Support Vector Regressor, Decision Tree, and K-Nearest Neighbor, with pixel counts as the input parameter for predicting pineapple weight. The outcome showed that YOLOv8 outperforms other YOLO models such as YOLOv7, YOLOv11 and RT-DETR achieving 99.8% Precision, 97.7% mean average precision, and 93.1% Recall for pineapple detection. For classification, this work has compared YOLOv8 with ML techniques, the result shows that YOLOv8 achieves better precision scores of 98% for unripe, 92% for semi-ripe, and 98% for ripe. For size estimation, this work has used BB and masking techniques, where the size error between measured and estimated (pixel to millimeter) pineapples were analyzed in terms of Root Mean Square Error, Mean Average Error, and Mean Absolute Percentage Error (MAPE). The MAPE value was below 6% for non-occluded pineapples when their size was estimated using EF and for occluded apples, it was 8% when the size was estimated using CE. The K-Nearest Neighbor model predicts pineapple weight with the highest accuracy, achieving an R² of 0.978 and RMSE of 4.12&#xa0;g. In this work, we show that our integrated approach of YOLOv8-based detection and classification, binary mask-based size estimation, and machine learning-based weight prediction provides an effective framework for automated pineapple monitoring in precision agriculture.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated pineapple detection, ripeness classification, and weight prediction using YOLOv8 and drone imagery

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Bidyarani Langpoklakpam,
  • Lithungo K. Murry,
  • R. Kumar

摘要

The precise detection and classification of pineapples at various stages of maturity are crucial for early yield estimation in agriculture. This study presents a framework utilizing YOLOv8 for pineapple detection and ripeness classification through drone imagery. After detecting and classifying pineapples, we employ bounding box and binary mask fitting techniques, including Circle Fitting (CF), Ellipse Fitting (EF), Circle Enclosing (CE), and Rotated Rectangle (RR), to estimate pineapple size by converting pixel-based measurements to millimeters. Finally, the current research predicted the weight of the pineapples by employing linear and nonlinear models. This work used linear models such as Linear Regression, Polynomial Regression, and nonlinear-based machine learning models such as Support Vector Regressor, Decision Tree, and K-Nearest Neighbor, with pixel counts as the input parameter for predicting pineapple weight. The outcome showed that YOLOv8 outperforms other YOLO models such as YOLOv7, YOLOv11 and RT-DETR achieving 99.8% Precision, 97.7% mean average precision, and 93.1% Recall for pineapple detection. For classification, this work has compared YOLOv8 with ML techniques, the result shows that YOLOv8 achieves better precision scores of 98% for unripe, 92% for semi-ripe, and 98% for ripe. For size estimation, this work has used BB and masking techniques, where the size error between measured and estimated (pixel to millimeter) pineapples were analyzed in terms of Root Mean Square Error, Mean Average Error, and Mean Absolute Percentage Error (MAPE). The MAPE value was below 6% for non-occluded pineapples when their size was estimated using EF and for occluded apples, it was 8% when the size was estimated using CE. The K-Nearest Neighbor model predicts pineapple weight with the highest accuracy, achieving an R² of 0.978 and RMSE of 4.12 g. In this work, we show that our integrated approach of YOLOv8-based detection and classification, binary mask-based size estimation, and machine learning-based weight prediction provides an effective framework for automated pineapple monitoring in precision agriculture.