STUDY ON DETECTION TECHNOLOGY OF DELINTED COTTONSEEDS VIGOR BASED ON HYPERSPECTRAL IMAGE TECHNOLOGY

Abstract. In order to seek a rapid detection of delinted cottonseed vigor without destroying, hyperspectral imaging technology was investigated to detect the delinted cottonseed vigor in this study. Experiment with different aging degree of three varieties of delinted cottonseeds as the research object, such as Xin Luzao 50, Xin Luzao 52, Xin Luzao 62. Firstly, the hyperspectral image of delinted cottonseeds in the range of 450-1013nm was collected. Secondly, we extracted three varieties of 720 RIOs and each region of interest contain a single entire region delinted cottonseeds. Thirdly, the extraction of spectral data used Savitzky-Golay smoothing method, combined with Mutiplicative Signal Correction (MSC) to cope with the noise. Finally, the ten diagnostic PC and special bands were respectively established a Discriminat analysis and Support vector machine (SVM). The results indicated that Discriminat analysis model about three varieties of delinted cottonseeds based on ten PCs applied as input is superior to Support vector machine model, the optimal classification aaccuracy of the discriminant model test can reach 100%. The performance of Discriminat analysis models built based on special wavelengths, the testing set accuracy rate of the three varieties about Xin Luzao 50, Xin Luzao 52, Xin Luzao 62 reached 98.3%, 98.3%, 96.6% respectively. The highest classification accuracy of the model established by the principal component is slightly higher than that of the characteristic band. This paper studied applying the hyperspectral image technology to detect the vigor of delinted cottonseeds, not only provides a new method for detecting delinted cottonseed vigor, but also lays a theoretical foundation for other seed vigor test.