MARTÍNEZ ESTUDILLO, FRANCISCO JOSÉ, ROMAINE, JAMES BRIAN, Johnson, P. , NIETO CARDONA, JUAN DIEGO, MILLÁN GATA, PABLO
No
Automatic Chard Area Detection
Conference Paper
Científica
01/01/2025
2-s2.0-105022249722
This work combines three traditional approaches: segmentation, detection, and classification, and was evaluated to automatically identify the vegetative area of chard in images. A technique was applied to enable farmers to efficiently detect vegetative regions and assess the overall health of the crop, reducing the subjectivity associated with manual inspections. The proposed deep learning model integrates the YOLOv11 object detector, K-means clustering, and the superpixel segmentation method. To optimise the labelling process, bounding boxes were used instead of detailed contour labels, significantly reducing the time and effort required for data preparation. The technique, referred to as YKMS, simplifies model training and facilitates precise identification of the vegetative area of chard, improving both efficiency and accuracy. The model was trained using 138 images captured in a real-world chard field. As a reference, YOLOv11 achieved recall, precision, and mAP values of 99.7%, 97%, and 77.2%, respectively. © 2025 IEEE.
Agriculture; Deep learning; Image segmentation; K-means clustering; Labels; Learning systems; Object recognition; Robotics; Superpixels; Area detection; K-means; Manual inspection; Objects detection; Segmentation; Super pixels; Threshold; Traditional approaches; Traditional approachs; YOLOv11; Computer vision; Object detection