Research On Wheat Hardness Detection Based On Near Infrared Spectroscopy

Wheat hardness is an important factor for quality classification and is closely related to the breeding of wheat.In order to satisfy the need to test the wheat hardness quickly,this subject will use “Yumai”as the sample.The first step is the preprocessing of the spectrum of wheat,then is the extraction of some specific wavelengths from the spectrum.Finally,based on the established wheat hardness prediction model,fast and non-destructive testing of unknown wheat samples hardness can be achieved.The detailed content of this subject is as follows:1.Preprocessing of the Near Infrared SpectralThe appearance of abnormal samples is inevitable due to factors such as the samples themselves and the collection techniques.Since the existence of abnormal samples may seriously affect the prediction ability of the model and the accuracy of the model evaluation,this subject is going to identify wheat spectral data based on Monte Carlo Cross-Validation pattern to eliminate the abnormal samples.In order to obtain both representative prediction and calibration samples,this subject employs the set partition method on wheat spectral data based on the spectral physicochemical value symbiotic distance method to obtain the prediction set samples.In order to eliminate irrelevant information such as high-frequency noise,baseline drift,sample background contained in the acquired spectral data,this subject uses a first-order derivative combined with multiplicative scatter correction to pre-process the spectral data,reducing or even eliminating the impact of the non-target factors on the model.2.Extraction of the Characteristic Wavelength of the Near Infrared SpectralIn order to deal with the problems that the pre-processed full-spectrum data has high dimensions,multiple wavelength information variables,and a large amount of calculation,the subject is based on the reverse interval partial least squares combined with a competitive self-adapting weighting algorithm to select useful wavelength variables,while reducing useless information to be introduced into the model which improved the stability and predictability of the prediction model.3.Construction of the predicted model for wheat hardnessThis project constructs wheat spectrum models based on the pre-processed and the extracted characteristic wavelengths,establishes linear partial least squares and nonlinear BP neural network prediction models.By comparison and analysis,the linear partial least squares model is selected as the final prediction model for testing wheat hardness.