Detection method of hazelnut mildew based on hyperspectral feature fusion
To realize the rapid and non-destructive detection of hazelnut mildew,the study fused spectral features with texture fea-tures,combining machine learning algorithms to establish a hazelnut mildew detection model.The hyperspectral images of hazelnut samples in the range of 400-1 000 nm were collected.The original spectrums were preprocessed with the standard normal variable transformation method.Dung beetle optimizer algorithm,particle swarm optimization algorithm,and successive projections algorithm were adopted to se-lect characteristic wavelengths.The hyperspectral images were reduced the dimensionality with principal component analysis method,and the optimal principal component images of the samples were selected according to the contribution size of the images,utilizing the gray-lev-el co-occurrence matrix method to extract five texture feature parameters on the four angles of samples.The hazelnut mildew detection K-nearest neighbor model was built based on spectral features,texture features,spectral features combined with texture features.Experimen-tal results indicated that the best model was the K-nearest neighbor model under the fusion of texture features and spectral features selected by dung beetle optimizer algorithm.The accuracy of the model training set and test set were 99.20％and 98.34％respectively,realizing rapid and non-destructive detection of hazelnut mildew.
万方数据
维普
