Abstract: 【Objective】Traditional chemical methods for assessing the storage quality of kiwifruit typically involve complex procedures and high costs, which may hinder their widespread use. These conventional approaches often require significant labor, time, and expensive reagents, making them less feasible for large-scale or routine quality control applications. Additionally, these methods usually result in the destruction of the fruit samples, which is not ideal for continuous monitoring. The complexity and cost associated with traditional chemical methods create barriers for smaller producers and can lead to inconsistencies in quality control across the industry. To address these issues, we propose a non-destructive testing method based on hyperspectral technology. This study aims to develop a reliable and efficient method for evaluating the quality of kiwifruit during storage without damaging the samples, thereby providing a more practical and economical solution for the kiwifruit industry. 【Methods】In this study, 110 ' Miliang-1' kiwifruit samples were used as experimental subjects. These samples were selected to represent a broad range of storage conditions and potential quality variations, ensuring that the findings of the study would be widely applicable. During the research, a hyperspectral imaging system was used to collect hyperspectral reflectance data of these kiwifruits at different storage times. This data collection included indicators such as titratable acidity, firmness, and soluble solid content, which are critical factors in determining the fruit's overall quality. Hyperspectral imaging technology can capture detailed spectral information across a wide range of wavelengths, providing rich spectral data that offers insights into the internal and external properties of the fruit. This non-invasive method enables the assessment of quality attributes without compromising the integrity of the samples, allowing for repeated measurements over time. However, to ensure the accuracy and reliability of the data, multiple preprocessing methods were employed to process the collected data. These preprocessing methods not only enhance signal quality but also effectively remove noise from the data, ensuring the precision and effectiveness of subsequent analyses. The preprocessing methods used in the study included Standard Normal Variate transformation (SNV), Multiplicative Scatter Correction (MSC), first-order derivative (1st-D), second-order derivative (2nd-D), and Savitzky-Golay smoothing (SG). These methods help correct baseline variations, scatter effects, and noise, improving the quality of the spectral data. Each preprocessing technique addresses specific issues within the spectral data, such as correcting for light scattering, baseline shifts, and other interferences, thereby optimizing the data for further analysis. To select the optimal hyperspectral wavelengths for predicting kiwifruit quality, Genetic Algorithm (GA) and Random Frog (RF) methods were employed. These algorithms are powerful tools for feature selection, capable of identifying the most informative wavelengths from the hyperspectral data. By pinpointing the most relevant wavelengths, these methods reduce the dimensionality of the data and enhance the efficiency of the predictive models. The selected wavelengths were then used to construct regression prediction models for kiwifruit quality indicators, including soluble solid content (SSC), firmness, and titratable acidity. The regression models utilized a combination of Support Vector Regression (SVR) and Backpropagation Neural Network (BP) algorithms to determine the optimal predictive performance for each quality indicator. These models are particularly suitable for handling the complexity and non-linearity of hyperspectral data, as they can effectively learn from the intricate relationships within the data. 【Results】The study found that the combination of preprocessing and wavelength selection significantly impacted the accuracy of the prediction models. For soluble solid content, the best model was 1st-D + GA-BP, with a coefficient of determination (R²) of 0.903 and a root mean square error (RMSE) of 1.731, indicating high accuracy in predicting kiwifruit SSC, reflecting the potential relationship between spectral data and SSC. For firmness, the best prediction model was 1st-D + RF-BP, with an R² of 0.900 and an RMSE of 0.879, demonstrating reliable predictive capability and highlighting the robustness of the model. For titratable acidity, the best model was 1st-D + GA-BP, with an R² of 0.857 and an RMSE of 0.225, showing good performance in predicting acidity levels and demonstrating the model's effective generalization to new data. These results underscore the effectiveness of the developed models and the significant role of preprocessing and feature selection in enhancing model performance. 【Conclusion】The successful application of hyperspectral technology in this study highlights its potential for non-destructive quality assessment of kiwifruit. By accurately predicting key quality attributes such as SSC, firmness, and acidity, hyperspectral imaging provides a powerful alternative to traditional chemical methods. This technology not only simplifies the assessment process but also reduces costs and preserves the integrity of the fruit samples. Additionally, it allows for continuous monitoring of fruit quality during storage, enabling timely interventions to maintain optimal conditions and prevent spoilage. This advancement could revolutionize quality control in the kiwifruit industry, providing a more efficient, cost-effective, and sustainable approach to maintaining high standards of fruit quality. The ability to monitor quality in a non-destructive manner also opens up new possibilities for research and development in the field of agricultural sciences, potentially leading to further innovations and improvements in fruit quality assessment and management.
PDF （）