<p>Color-changing melons exhibit a prolonged growth cycle and variable ripening periods, complicating their harvesting. Consequently, detection models have emerged as an effective solution to assist in harvesting. However, current object detection models for assessing the ripeness of color-changing melons face challenges, including low accuracy, slow processing speed, and large model sizes. To address these limitations, this study introduces a lightweight detection model, FHG-DETR. The model employs FasterNet as the feature extraction network to reduce the number of parameters and computational load while enhancing feature extraction capabilities and increasing detection speed. Additionally, the integration of the HiLo Attention module improves the model’s capacity to focus on details and distinguish the target from backgrounds effectively. Furthermore, the model incorporates a novel BiFPN-GSDI feature fusion network. This uses BiFPN as the core structure, with the GSDI module optimizing feature fusion, thereby enhancing detection accuracy and reducing computational requirements. The experimental results demonstrate that FHG-DETR achieves a 4.6% improvement in mAP50 compared to RT-DETR in ripeness detection of color-changing melons, with the GSDI module contributing significantly to this enhancement. Moreover, FHG-DETR reduces the number of parameters by 59% and computational load by 65.6% compared to RT-DETR, while increasing detection speed by 23.4% to 246.6 FPS. These findings indicate that FHG-DETR enables more accurate and efficient detection of color-changing melons across various ripeness stages.</p>