|Peng, Yankun - CHINA AGRICULTURAL UNIV|
Submitted to: Proceedings of SPIE
Publication Type: Proceedings
Publication Acceptance Date: September 10, 2007
Publication Date: October 12, 2007
Citation: Peng, Y., Lu, R. 2007. A recursive method for updating apple firmness prediction models based on spectral scattering images. Proceedings of SPIE. 6761:67610U. Interpretive Summary: Spectral scattering is a new technique that measures light scattering characteristics of fruit at multiple wavelengths. Since light scattering is related to the physical and structural properties of the fruit, it can be used for assessing the firmness of apples, which is an important parameter in evaluation of their overall quality grade and consumer acceptance. The spectral scattering technique, however, relies on proper selection of samples for developing a calibration model. The calibration model developed for a given population of fruit may not be suitable for another population that is different in harvest time and/or postharvest handling/storage history. Hence, it is necessary to develop an effective method to update the calibration model so that it can be used for apples of different harvest times and postharvest storage histories. In this research, a recursive method of updating calibration models was proposed and validated for predicting the firmness of apples from two different storage conditions. Four different methods of selecting new calibration samples were examined and compared. Results showed that the proposed model-updating method, coupled with an appropriate sampling technique, could effectively predict the firmness of apples from a new population. The method required fewer samples for updating the calibration model to achieve acceptable firmness prediction results. Hence it may provide an economic way of implementing the spectral scattering technique for assessing apple firmness. With further testing on a broader range of apple samples, the model-updating method can facilitate the application of the spectral scattering technique in sorting and grading apples for firmness, which is critical to ensure the overall quality and consistency of fresh fruit delivered to the marketplace.
Technical Abstract: Multispectral scattering is effective for nondestructive prediction of fruit firmness. However, the established prediction models for multispectral scattering are variety specific and may not perform appropriately for fruit harvested from different orchards or at different times. In this research, a recursive least squares method was proposed to update the existing prediction model by adding samples from a new population to assure good performance of the model for predicting fruit firmness from the new population. Multispectral scattering images acquired by a multispectral imaging system from ‘Golden Delicious’ apples that were harvested at the same time but had different postharvest storage time periods were used in the development of the updating method. Radial scattering profiles were described by a modified Lorentzian distribution (MLD) function with four profile parameters for eight wavelengths. Multi-linear regression was performed on MLD parameters to establish prediction models for fruit firmness for each group. The prediction model established in the first group was then updated by using selected samples from the second group, and four different sampling methods were compared and validated with the rest apples. The prediction model corrected by the model-updating method gave good firmness predictions with the correlation coefficient (r) of 0.86 and the standard error of prediction (SEP) of 6.11 N. This model updating method is promising for implementing the spectral scattering technique for real-time assessment of apple fruit firmness.