Abstract Mango (Mangifera indica L.) is a major tropical fruit that can develop internal physiological disorders at late ripening stages. These include jelly seed characterized by a transparent and jelly tissue around the seed that eventually becomes a brown ring enclosing the seed, and black flesh characterized by a diffuse brown discoloration that covers the seed. Both disorders can result in high postharvest losses due to the fact that little information is available about mechanisms involved and efficient control approaches. The objective of this study was to establish the feasibility of using a visible and near-infrared (Vis-NIR) portable spectrometer for predicting at harvest and detecting mangoes with internal disorders, such as jelly seed and black flesh after storage. A total of 141 'Keitt' mangoes from two commercial harvests were measured spectrally between 400–1100 nm on two opposite cheeks, at harvest and after 30 d at 12 °C. Spectra data and the incidence of jelly seed and black flesh after storage were used to develop classification models using logistic, linear discriminative analyses (LDA), supporting vector machine, functional data and random forest modeling approaches. The results show that wavelengths between 550 and 650 nm can be used to predict at harvest and detect after storage, fruit with internal physiological disorders, such as jelly seed and black flesh. However, it was not possible to differentiate internal disorders from each other. The spectral data show that healthy fruit have higher reflectance intensity than jelly seed and black flesh ones, both at harvest and after storage. The best classification models were obtained with Logistic and LDA model development approaches. In the validation process for internal disorder prediction at harvest, the Logistic model showed accuracy of 65 %, sensitivity of 78 % and specificity of 49 %, whereas the LDA model showed accuracy of 63 %, sensitivity of 76 % and specificity of 46 %. In the validation process for detecting internal disorders after storage, the Logistic model showed accuracy of 71 %, sensitivity of 75 % and specificity of 67 %, whereas the LDA model showed accuracy of 76 %, sensitivity of 78 % and specificity of 73 %. In conclusion, Vis-NIR technology associated with Logistic and LDA modeling approaches can be used to predict at harvest and detect after storage the incidence of jelly seed and black flesh in mangoes.

中文翻译：

---

使用手持式 Vis-NIR 光谱仪无损预测和检测“Keitt”芒果内部生理障碍

摘要 芒果（Mangifera indica L.）是一种主要的热带水果，在成熟后期会发生内部生理紊乱。这些包括果冻种子，其特征是种子周围有透明的果冻组织，最终变成包围种子的棕色环，以及黑色果肉，其特征是覆盖种子的弥漫性棕色变色。由于有关机制和有效控制方法的信息很少，这两种疾病都可能导致高收获后损失。本研究的目的是确定使用可见光和近红外 (Vis-NIR) 便携式光谱仪在收获时预测和检测芒果内部紊乱的可行性，例如储存后的果冻种子和黑色果肉。共有 141 个“凯特” 在收获时和 12°C 下 30 天后，对来自两个商业收获的芒果在两个相对的脸颊上进行了 400-1100 nm 的光谱测量。光谱数据以及储存后果冻种子和黑肉的发生率用于使用逻辑、线性判别分析 (LDA)、支持向量机、函数数据和随机森林建模方法开发分类模型。结果表明，550-650 nm波长可用于预测收获时和储存后检测内部生理紊乱的水果，如果冻种子和黑色果肉。但是，无法区分内部疾病。光谱数据表明，健康水果在收获和储存后都比果冻种子和黑色果肉具有更高的反射强度。最好的分类模型是通过 Logistic 和 LDA 模型开发方法获得的。在收获时内部紊乱预测的验证过程中，Logistic模型的准确率为65%，敏感性为78%，特异性为49%，而LDA模型的准确率为63%，敏感性为76%，特异性为46%。在存储后检测内部疾病的验证过程中，Logistic模型的准确率为71%，敏感性为75%，特异性为67%，而LDA模型的准确率为76%，敏感性为78%，特异性为73%。总之，与 Logistic 和 LDA 建模方法相关的 Vis-NIR 技术可用于预测收获时和储存后检测芒果中果冻种子和黑肉的发生率。