<p>Fruit Quality Estimation (FQE) is essential for assessing the freshness, nutritional value, and overall quality of fruits. Hyperspectral Imaging (HSI) provides a nondestructive and accurate approach for this task. However, existing methods rarely integrate multiple indices such as the Reflectance Entropy Index, MalonDialdehyde Index, and Red Edge Index, which are critical for comprehensive quality assessment. To address this limitation, this study proposes a novel framework that combines advanced preprocessing, feature optimization, and deep-learning techniques for precise FQE. The HSI data of apples and oranges were first collected and pre-processed, with the scattering effects removed using Entropy Function-based Multiplicative Scatter Correction. Fruit detection was performed using the Gaussian Weighted-You Only Look Once technique, followed by surface reconstruction via the Multi-Chart Implicit-based Geometry Smooth Neural Implicit Surface method. Physical, nutritional, and defect-related indices were then calculated and fruit quality was labeled using a pii-shaped adaptive network-based fuzzy inference system. Relevant features are extracted, and optimal features are selected through Sinusoidal Quadratic Function-based Fennec Fox Optimization. Finally, fruit quality is estimated using the Drop Connect Parameterized Rectified Linear Unit Residual Network classifier. The proposed framework achieved an accuracy of 99.11% and a precision of 99.17%, outperforming existing approaches, and demonstrated the effectiveness of integrating HSI with advanced deep learning and optimization techniques for non-destructive and reliable fruit quality assessment.</p>