<p>Maturity assessment plays an important role in precision agriculture. This work proposes a fine-grained analysis method for fruit coloration information and obtain a maturity index (MI) for quantifying the fruit growth-maturation process. In the preprocessing of the dataset, FreqViT was proposed as a fruit segmentation model optimized for capturing edge information and hierarchical relationships in small-scale image inputs. A training-validation strategy based on adaptive weight selection was introduced to maximize the advantages of FreqViT in detail processing. Furthermore, the Hue channel in the HSV color space was adopted to represent and analyse fruit coloration information. A process for constructing a Valid Mask was also proposed, and the large mask inpainting (LaMa) was applied for image inpainting to address overexposed regions. Finally, maturity was quantified by calculating the difference between pixel hue values and the standard maturity. The efficiency, accuracy, and versatility of the model were validated through targeted analysis and batch detection. In complex orchard environments, FreqViT-S / L / X achieves 93.4% / 94.3% / 94.5% Top-1 mIoU, respectively. The training strategy, color space, and image inpainting algorithm achieves optimal processing performance in comparative evaluations. The maturity quantification not only provides the digital representation but also outperforms classification in Precision, Recall, and F1-score. This work facilitates intelligent and refined orchard production, supporting more scientific orchard management and harvesting decisions.</p>