Research On Nondestructive Detection Of Cucumber Seeds' Quality Based On Hyperspectral Image Technology

Cucumber is one of the largest planting vegetables in China.The quality of cucumber seeds directly affects the germination rate and yield after sowing.At present,the methods for detecting the quality of cucumber seeds have shortcoming such as long detection time,low efficiency,and damage to seeds.With the rapid development of hyperspectral image technology,hyperspectral image technology has been used in the agricultural field more and more widely.In this paper,cucumber seeds were used as the research object.And a non-destructive detection method on the vigor,variety and water content of cucumber seeds based on hyperspectral image technology is proposed.The main contents are as follows:Firstly,hyperspectral image technology was used to test cucumber seeds of different vitality.The cucumber seeds were artificially aged by high temperature and high humidity method,and divided into three gradients: without aging,aging for 36 hours,and aging for 72 hours.A hyperspectral image acquisition system was used to acquire hyperspectral images of three gradient cucumber seeds.After the collection,some seeds were tested for catalase activity according to GB/T5522-2008.Some seeds were tested for germination rate according to GB/T5520-2011,and the vigor characteristics of cucumber seeds were verified.The region of interest was identified and hyperspectral data was extracted by ENVI software.The acquired hyperspectral data was preprocessed,and different pretreatment methods(S-G,MSC,SNV)were compared.The results showed that SNV was the best.Then using principal component analysis and successive projections algorithm to extract features,and establish ELM and SVM models respectively.Compared with the models,the optimal model is SPASVM model(radial basis function kernel).The correct rate of test set classification is 98.6%,and the correct rate of cross-validation is 95.3%.Therefore,it is feasible to use hyperspectral image technology to identify different vital cucumber seeds.Secondly,three different varieties of cucumber seeds were identified by hyperspectral image technique.Hyperspectral imagery was used to obtain hyperspectral images of seeds of "Xinjinchun No.4","Zhaibubai" and "Fuyang No.35".The region of interest was determined and hyperspectral data was extracted by ENVI software.Compared different preprocessing methods,selecting the optimal preprocessing method,then using principal component analysis and successive projections algorithm to reduce feature dimensionality.Established ELM and SVM models,and the best result is SPA-SVM model(radial basis function kernel).The correct rate of the test set was 96%,and the correct rate of cross-validation was 91.6%.Thirdly,the moisture content of cucumber seeds was quantitatively detected by hyperspectral image technique.The hyperspectral image of cucumber seeds in the range of 840-1766 nm was acquired by hyperspectral image acquisition system,and the hyperspectral data in the region of interest was extracted by ENVI.According to the national standard GB/T3543.6-1995,the water content was measured by the 105 °C constant weight method.The competitive adaptive weighting algorithm and successive projections algorithm are used to extract the spectral features,and then the detection model was established by PLSR.The SPA-PLSR model has the highest prediction accuracy.The Rp2 of the prediction set is 0.85,the RMSEP is 1.79%,and the crossvalidated RCV2 is 0.76.The RMSECV is 1.66%.Based on MATLAB and Visual Studio,the cucumber seed quality testing software was developed to realize the function of detecting seed quality.The software based on MATLAB R2016 a can realize the reading of hyperspectral data of cucumber seeds,data preprocessing,characteristic wavelength selection and modeling functions,and basically realize the classification detection of different vital cucumber seeds.The integrated software based on Visual Studio 2013 is written in C# language,and the model built by MATLAB is compiled into dll file by COM technology,which realizes the embedding of the algorithm.Using the software,data reading,pre-processing,and characteristic wavelength selection can be realized quickly and easily,and the seed vitality classification and moisture content detection can be realized,which basically meets the requirements of cucumber seed detection.