首页 > , Vol. , Issue () : -
As an important grain crop in China, the maize has many varieties and is prone to misclassification, affecting agricultural security and food production. With the development of the hyperspectral image and deep learning technology, it is possible to identify crop varieties using the combination of both of them. Convolutional neural network (CNN), the most representative algorithm of deep learning technology to deal with the image classification task, needs a large number of training samples in the process of the training model. However, it is difficult and time-consuming to obtain a large number of hyperspectral images of maize seed samples. Aiming at the problem of the huge number of modeling samples required for the traditional method based on CNN for crop identification of hyperspectral image, the variety identification model of maize seeds based on hyperspectral pixel-level spectral information and CNN is proposed. First of all, hyperspectral images of different varieties of maize seeds in the range of 400~1000nm were obtained, 203-dimensionality spectral information of all the pixels of samples was extracted. However, the huge amount of spectral information creates the problem of dimensional disaster and greatly increases the computational cost. Second, to reduce dimensionalities of the sample spectral information, the principal component analysis (PCA) algorithm was used to reduce the spectral dimension to 8 dimensions, which effectively shortens the operation time. Third, the pixel-level spectral information of the sample (i.e., the spectral information of all the pixels of the sample) is applied to the support vector machine (SVM) and the K-nearest neighbor classification (KNN) models, in addition to the CNN model. The experimental results demonstrate that for CNN, SVM and KNN recognition algorithms, based on pixel-level spectral information models show a more stable and efficient recognition effect than the seed-level one. (i.e., the average of all pixel spectral information of each sample). This is because based on the seed-level information model does not make full use of the sample pixel spectrum and spatial information of pixels, which needs a large number of modeling samples. When the number of samples used to build the classification model is the same, the CNN model has a significantly better recognition effect than the SVM and KNN models. According to all the pixel-level classification results, a majority voting strategy was used to identify the corn seed sample variety and the sample recognition accuracy was up to 100% (Note: 100% refers to the identification accuracy when the number of samples of modeling set and testing set is 0.27 and 0.32. As the number of samples of the testing set increases, the identification accuracy will decrease.) Last but not least, the t-distribution random neighbor embedding algorithm (t-SNE) was used to realize the visualization of output eigenvalues of CNN, the features of different maize seed varieties are clearly bounded in the visualization, which adequately verifies the validity of the species recognition model based on hyperspectral pixel-level information and CNN. In the rare case of modeling seed samples, the non-destructive and efficient variety identification of the maize seed is realized, which will provide a theoretical basis for precision agriculture.