转基因番茄鉴别模型维护方法

Calibration model maintenance method for transgenic tomato discriminant

利用光谱数据建立的校正模型存在着模型适用性差的问题,当这个模型已无法预测新样本或预测偏差较大的时候,就需要对模型进行维护.利用建立的转基因番茄鉴别模型,尝试用模型更新和吸光度值修正2种方法,对同一批次相同成熟度、不同批次相同成熟度及同一批次不同成熟度的样本进行预测并进行模型维护.结果表明:校正模型可用于预测同一批次相同成熟度的转基因番茄及其亲本;模型更新法对不同批次相同成熟度的转基因番茄样本预测效果较好,而吸光度值修正法对亲本样本预测效果更好;在预测同一批次不同成熟度的样本时,用吸光度值修正的方法更优于模型更新法.

To improve the model applicability of calibration model built from spectroscopy, the model maintenance is needed when the prediction precision of original model is poor for predicting new samples. The discriminant analysis （DA） was used to build transgenic tomato parent sample discriminant calibra-tion model to predict sample ripeness with same batch at same ripeness stage, different batches at same ripeness stage and different batches at different ripeness stages. To improve model applicability, model update by adding new typical samples into original model and absorbance modification were adopted. The results indicate that calibration model can be employed to predict samples with same batch at same ripe-ness stage. The prediction performance of transgenic samples with different batches at same ripeness stage can be improved by model update, and the prediction performance of parent samples can be improved by absorbance modification. For predicting transgenic tomatoes and parent samples with same batch at diffe-rent ripeness stage, the efficiency of absorbance modification is better than that of model update.