Methods Of Soft And Hard Identification And Moisture Prediction Of Single Kernel Wheat Based On Hyperspectral

Wheat is one of the most important crops in China,which plays an important role in modern agriculture.The inheritance of wheat hardness is relatively simple,and early generation selection is effective.At the same time,there are great differences between soft wheat and hard wheat in particle size,moisture absorption capacity and other aspects,resulting in different uses.In addition,when the moisture of single wheat kernels deviates from the normal value,it will cause a series of reactions.High-moisture wheat kernels are prone to freeze damage and reduce the germination rate.If not handled in time,it will cause a lot of losses.In this paper,the hyperspectral technology is applied to the soft and hard identification and moisture detection of single wheat kernel,which provides a rapid identification methods for selecting high-quality wheat kernels.The main research and conclusions are as follows:1.Based on hyperspectral imaging technology combined with chemometric methods,it is applied to the identification of single kernel soft and hard wheat.A single kernel soft and hard wheat identification model based on SVM,KNN,CNN and CNN transfer learning was established.A total of 50 samples of wheat from different regions,varieties and harvest years at home and abroad were collected.2500 samples of wheat kernels were selected and hyperspectral images were collected.Using a physical property analyzer to measure the hardness of wheat samples,including 781 soft wheat,113 mixed wheat and 1606 hard wheat.To explore the influence of different chemometric methods on the soft and hard identification model of single wheat kernel.The accuracy of the model is 83.21%based on SVM and 85.21%based on KNN.At the same time,a convolution neural network structure is designed for the recognition of the hyperspectral pseudo-color image of single wheat kernel,establish the qualitative recognition model of single kernel soft and hard wheat.The model has an accuracy rate of 88.50%.Convolutional neural network transfer learning is also used to achieve the identification of the hardness wheat kernels based on hyperspectral images,and finally the accuracy rate on the verification set is increased to 94.00%,and the accurate rate on the test set is 92.80%.This effectively completes the identification of the soft and hard wheat kernels.The results show that the accurate and stable detection results can be obtained by using the hyperspectral method to identify the pseudo-color image of single wheat kernels,it is feasible to identify the soft and hard of single wheat kernels by hyperspectral image.2.Quantitative prediction of moisture content of single wheat kernel by partial least square regression based on hyperspectral.The wheat kernels were divided into crease up(recorded as A-plane)and crease down(recorded as B-plane)for spectral scanning.Four prediction models,MA,MB,MC and MD,are constructed by using the spectral data of A-plane and B-plane.The model MA and model MB are composed of spectral data of A-plane and B-plane,respectively.The model MC consists of the average spectrum of A-plane and B-plane of each kernel,and the model MD was formed by the spectral data of A-plane and B-plane together.The partial least squares regression(PLSR)was explored to predict the moisture content in single wheat kernel.The experimental results show that the order of the prediction performance of the four models from high to low is model MD,model MC,model MA and model MB.That is to say,model MD has the best prediction effect,because the model MD has better robustness when mixing the spectral data of A-plane and B-plane.For model MD,after pretreatment by First Derivatives and Standard Normal Variate,the RMSEC is 1.20%,R_C is 0.93,RMSEP is 1.36%,R_P is 0.90.The results show that it is feasible to detect the moisture of single wheat kernels by hyperspectral.Using the established model MD based on PLSR,combined with hyperspectral imaging technology to obtain the moisture distribution map of single wheat kernel,the whole disc of single wheat kernel moisture visualization is realized.The results show that it is feasible to use hyperspectral technology to the soft and hard identification and moisture prediction of single kernel wheat.In addition,the use of hyperspectral imaging technology combined with the established PLSR moisture prediction model to draw a wheat kernel moisture distribution map to visualize the moisture of single wheat kernel.Through the moisture distribution map of single wheat kernel,we can locate the kernels with abnormal moisture content,which provides a reference for the future research on online detection and automatic removal of single wheat kernel with abnormal water moisture.