Conventional chemical methods to detect pesticide residue are complex, and need professional instruments as well as much time. NIR spectral analysis provides an effective way for pesticide residue detection because of simple operation, rapid analysis and non-destruction of samples. However, conventional calibration models are only effective after spectra were measured, and different models are needed for different instruments. In this study, we propose a novel calibration transfer method by using sequential forward selection to transfer calibration models between different crops and instruments. The calibration model built by master instrument can identify three kinds of pesticide including Chlorothalonil, Chlorpyrifos and Buprofezin. Spectra obtained by slave instrument can also be predicted by the models with the method. The experiment results show that the prediction accuracies increased from 50% up to 80% by using our method.
In recent years, the detection of pesticide residues is an important research field in China. Traditional methods which include biochemical methods and chromatography methods have high requirement for the experimental environment. Thus traditional methods cannot complete tasks in a short time. NIR spectrum represents physical features of matters and pesticide residues on fruit can change the NIR spectrum reflected by surface. It is feasible to make sure that if there is pesticide residues on fruit or not by analyzing the NIR spectrum reflected by surface. The method of spectrum analysis is easy to conduct and faster than traditional methods.