BlueSegamba-UNet: A Lightweight Super-Resolution-Aware Framework for UAV-Based Blueberry Ripeness Segmentation

Accurate blueberry ripeness assessment from UAV imagery is hampered by low spatial resolution, tiny fruit size, and heavy foliage occlusion. The study presents BlueSegamba-UNet, a framework integrating MambaIR-based super-resolution reconstruction with an improved CM-UNet. The network integrates three improvements to boost segmentation accuracy without adding noticeable latency. On our blueberry dataset, BlueSegamba-UNet attains 79.3% IoU for mature blueberries, 66.1% for immature blueberries, and an overall mIoU of 81.2%, surpassing CNN-, Transformer-, and Mamba-based baselines while keeping inference under 150ms. These results highlight the value of combining super-resolution and state-space segmentation for real-time, large-scale blueberry ripeness monitoring.