目的 将近红外无损检测技术应用于智能包装生产线上，以快速、准确地检测PE包装蓝莓的新鲜度。方法 以蓝莓含水率和可溶性固形物(SSC)为评价指标，使用SNV、MSC和DT结合UVE处理所获得的光谱数据，通过PLSR和RF建立散装蓝莓和PE保鲜膜包装蓝莓的含水率和SSC预测模型，通过比较校正决定系数Rc²、验证决定系数Rp²和均方根误差来验证模型的准确性。结果 建立散装蓝莓含水率预测模型，选择最佳预处理方法为SNV，使用PLSR建模，最佳主成分数为9，Rc²为0.971，Rp²为0.933;建立PE保鲜膜包装蓝莓含水率预测模型，选择最佳预处理方法为SNV，使用RF建模，Rc²为0.923，Rp²为0.876;建立散装蓝莓SSC预测模型，选择最佳预处理方法为DT结合UVE，使用RF建模，Rc²为0.942，Rp²为0.869;建立PE保鲜膜包装蓝莓SSC预测模型，选择最佳预处理方法为MSC结合UVE，使用PLSR建模，最佳主成分数为7，Rc²为0.849，Rp²为0.707。结论 通过对比散装蓝莓、PE包装蓝莓两者的预测模型，发现PE膜会影响预测模型的精度但不影响使用，为在智能包装生产线上快速无损检测蓝莓新鲜度提供了一种实用的方法。

The work aims to apply near infrared nondestructive testing technology to the intelligent packaging production line to detect the freshness of blueberries packaged with PE film accurately and rapidly. The moisture content and soluble solids content (SSC) of blueberries were taken as evaluation indexes. Furthermore methods of SNV, MSC and DT combined with UVE were used to obtain spectral data. Then PLSR and RF were used to build prediction models of moisture content and SSC of bulk and PE packaged blueberries. The accuracy of model was verified by comparing the determination coefficient and root mean square error. The prediction model of moisture content of bulk blueberries was established. SNV and PLSR were the best pre-processing and modeling methods, respectively. The best PCA was 9, Rc2 was 0.971, Rp2 was 0.933; For establishing a prediction model of moisture content of blueberries packaged with PE film, SNV and RF were the best pre-processing and modeling methods, respectively. Rc2 was 0.923, Rp2 was 0.876; For establishing a prediction model of SSC of bulk blueberry, DT combined with UVE, and RF were the best pre-processing and modeling methods, respectively. Rc2 was 0.942, Rp2 was 0.869; For establishing a prediction model of SSC of blueberries packaged with PE film, MSC combined with UVE, and PLSR were the best pre-processing and modeling methods, respectively. The best PCA was 7, Rc2 was 0.849, Rp2 was 0.707. By comparing the prediction models of bulk blueberries and PE packaged blueberries, it is found that PE film affects the accuracy of the prediction model, but doesn't affect the use of it. The work provides a practical method for rapid nondestructive testing of blueberry freshness in intelligent packaging production line.