A Hue-Based Segmentation for Melody Watermelon Images Before Harvesting

Abstract:

Melody watermelon, a hybrid variety valued for its uniform morphology and sweet flesh, has gained increasing importance in precision agriculture. Accurate fruit segmen-tation from on-field images is a fundamental prerequisite for downstream tasks such as yield estimation, maturity assessment, and sweetness identification. However, existing segmentation approaches, particularly deep learning and clustering-based methods, are often computationally intensive and less robust under complex field conditions. The study proposes a lightweight, hue-based segmentation framework using the HSV color model to isolate Melody watermelons from natural field backgrounds. The method retains pixels with hue values above a selected threshold and applies multi-stage fil-tering, including Gaussian, median, and averaging filters, followed by morphological refinement to enhance boundary delineation and suppress background noise such as soil and foliage. An ablation study is conducted to analyze the impact of different hue thresholds, leading to the selection of an optimal configuration. Segmentation perfor-mance is evaluated against ground truth annotations using standard metrics, including Dice Similarity Coefficient, Intersection over Union, Precision, Recall, F1-score, and Volumetric Overlap Error, achieving a Dice coefficient of up to 0.96. A comparative analysis further confirms the robustness and effectiveness of the proposed approach for real-field agricultural applications.