Research On Intelligent Detection Technology Of Hard Bean Seed

Snap beans not only contain abundant nutrients,but also have certain edible and medicinal value,favored by most consumers and growers.The quality of snap bean seeds directly determines the quality,yield,and planting efficiency of snap beans.However,the common problem of seed hardness in snap beans seriously affects their large-scale industrial cultivation and yield.How to quickly and accurately identify hard seeds before planting has become a focus of attention for snap bean growers and seed distributors.The traditional method of detecting seed hardness in snap beans is mainly the imbibition method,which has high detection costs and low accuracy,restricting the widespread development of hardness detection.Detection based on hyperspectral technology has the advantages of accuracy,speed,and non-destructiveness.This study combines hyperspectral imaging technology with machine learning technology to research a fast and accurate detection technology for snap bean seed hardness,achieving good results.(1)The balance of the sample dataset has been achieved.In this study,spectral values of seeds were obtained from high spectral images of individual seeds,and synthetic minority oversampling technique and Tomek Links were combined to achieve balance in the number of hard and non-hard seed samples.(2)The prediction of snap bean seed hardness was achieved based on HSI technology.First,after image processing operations,the average spectral curve of individual seeds was obtained.Then,with the combination of multiple preprocessing methods,the first-order differential(1D)spectral preprocessing method was determined as the optimal method based on the model detection accuracy and F1 value.Subsequently,20 feature bands were obtained from the 1D processed average spectrum using the successive projection algorithm(SPA).Finally,the intelligent detection of hard seeds was achieved by establishing a radial basis function support vector machine(RBF-SVM)model.The results showed that the accuracy of the full-band detection was 88.68%,and the detection accuracy of the feature bands was89.32%.(3)By analyzing the characteristic bands of active substances(such as proteins,fibers and proanthocyanidins)inside hard seeds,the correlation between characteristic wavelengths and seed hard characteristics was further verified.The validity of the extracted feature wavelengths is verified by physical property analysis.Through the determination of characteristic wavelengths,it was found that the internal substances related to hard seeds mainly included proanthocyanidins,lignin,tannins,cellulose and proteins.In this study,the indexes corresponding to the characteristic wavelengths extracted by the SPA method could reflect the contents of these five substances.Therefore,the extracted characteristic wavelengths and RBF-SVM can be used for hard detection of bean seeds.