High accuracy on-line estimation for fruit internal quality is still a challenge due to varying geometric structures and orientations. In this study, multiple full-transmittance spectra were collected using short-integration-time mode for each apple. The signal-to-noise ratio of each collected spectrum changed with the measurement position of the fruit due to the heterogeneity of internal composition. To explore the distribution character of transmittance spectra across the apple structure to guide the development of an on-line fruit quality determination system, a methodology which we called ‘zone combination modelling’ was proposed for selecting the most effective spectra for SSC prediction. The orientation of stem–calyx axis vertical was selected as the preferred orientation for quality prediction of ‘Fuji’ apple based on the analysis of the variation and quality of full-transmittance spectra. The most ineffective and most effective zone combinations for SSC prediction were determined by investigating the effect of transmittance spectra within different zone combinations on SSC prediction ability. Ten effective wavelengths selected from the most efficient zone combination were used to develop an optimal prediction model. Results showed that the contribution of different spectral measurement zones to SSC prediction capability varied and that in particular, those collected from the apple core zone should be removed when building SSC prediction models. The coefficient of determination and root mean square errors of prediction and validation sets of SSC, respectively, were 0.733 and 0.61%, 0.721 and 0.71% for the optimal model, indicating that zone combinations model was promising for SSC prediction of apple.

中文翻译：

---

基于全透射光谱法在线无损预测苹果可溶性固形物含量的最优分区组合模型

由于不同的几何结构和方向，水果内部质量的高精度在线估计仍然是一个挑战。在这项研究中，使用短积分时间模式为每个苹果收集了多个全透射光谱。由于内部成分的异质性，每个采集光谱的信噪比随着水果的测量位置而变化。为了探索整个苹果结构中透射光谱的分布特征以指导在线水果质量测定系统的开发，我们提出了一种我们称为“区域组合建模”的方法，用于选择最有效的光谱进行 SSC 预测。在对全透射光谱的变化和品质进行分析的基础上，选择茎-花萼轴垂直方向作为'富士'苹果品质预测的首选方向。通过研究不同区域组合内的透射光谱对 SSC 预测能力的影响，确定了 SSC 预测最无效和最有效的区域组合。从最有效的区域组合中选出的十个有效波长用于开发最佳预测模型。结果表明，不同光谱测量区对 SSC 预测能力的贡献各不相同，特别是在构建 SSC 预测模型时，应去除从苹果核心区采集的光谱。