论文部分内容阅读
The quality of rice will greatly change as the time has passed by because the inner physical and chemical components are apt to be affected by the exterior environment parameters such as temperature, humidity andillumination.Due to the price difference between the new and the old rice, the old rice were often polished and sold as the new rice.However, most of the people had difficult to distinguish them.The technology of fast, accurately and non-destructively detecting the old rice and new rice was needed to be explored.A new method based on the near infrared spectroscopy (NIRS) technique was presented for rapid and accurate identification of two kinds of the old rice and new rice.In our experiment, the old rice had been stored for more than two years and the new rice had been stored for less than half a year after the manufacture.The NIRS data was obtained by using a field spectroradiometer (FieldSpec 3 FR, Analytical Spectral Devices (ASD),Inc.USA) within wavelengths between 350 and 1070 nm with high resolution of 1 nm.The method of reflection measurement was employed to obtain NIRSof rice samples.Direct orthogonal signal correction preprocessing method (DOSC) was used to reduce the influence of light scattering,background noise, and baseline shift during experiment.Two classes of quantitative numerical statistical methods of Durbin-Watson test and Run test and theenhanced partial residual plot analysis method (APaRP) were respectively employed toestimate the nonlinearity of NIRSdata.It could be seen that there were obvious nonlinear structures in the obtained rice spectral data.Three kinds of non-linear manifold dimension reduction methods of isometric feature mapping (ISOMAP), local linear embedding (LLE) and laplacianeigenmaps (LE) were respectively used to reduce the dimensions of nonlinear frameworks of spectral data.The recognition precision computed by the partial least squares (PLS) algorithm was unsatisfied.It wasbecause that the measured reflectance NIRSof rice displayed the strong nonlinearity.The linear modeling method of PLS have difficult to discriminate the time attributions in the traditional linear space, which led to the accuracy decline.The manifold variables obtained by the nonlinear manifold dimension reduction methods were used as the input of kernel partial least squares (KPLS).The KPLS algorithm took advantage of the kernel function to map the processed nonlinear reflectance spectrum to the high-dimensional kernel space to alleviate the effects of nonlinearity of the components.A total of 400 rice samples were used for the experiment, each variety consisted of 200 samples.The rice samples were randomly divided into the training set and test set.The training set consisted of 350 samples while the testset consisted of 50 samples.The parameters of models were optimized using cross-validation method in terms of the forecast results.The ISOMAP-KPLS model obtained the highest prediction accuracy of 94.67%.It could be concluded that the NIRS techniques combined with the ISOMAP-KPLS method could be successfully used to discriminate the new rice and old rice.The complexity of the detection methodsbased on the NIRS wasmuch simpler than the traditional physical and chemical detection method.The proposed methods provided a new pathfor the future rapid nondestructive detection of the old and new rice.