Fruit firmness is a critical quality indicator influencing texture, processing, and resistance to post-harvest diseases. Spectroscopy is commonly applied for firmness assessment
however, limited sample sizes in target domains (datasets to be analyzed) often affect detection performance. To address this limitation, a spectral feature-enhanced method is developed to integrate spectral data from related domains (datasets with similar spectral characteristics). The method incorporates multi-scale spectral inputs, a joint domain feature extractor for shared features, and a target domain feature extractor for domain-specific features. Reconstruction mechanisms and similarity constraints are employed to ensure that the extracted features capture intrinsic domain characteristics. The combined features serve as inputs for firmness prediction. Evaluation using data from three apple varieties across 30 joint scenarios indicates that the proposed method outperforms single-scenario and existing approaches in 28 scenarios, demonstrating its effectiveness in managing data variability and enhancing prediction performance.