An MLP Network Based on Residual Learning for Rice Hyperspectral Data Classification
Xiaojiang Tang, Xin Liu, Pengfei Yan, Bao-xia Li, Haoyu Qi, Feng Huang
Abstract
In order to classify several kinds of rice (including the rice grown by plasma seed treatment), the datasets of the hyperspectral images (HSIs) of rice were constructed. Multilayer perceptron (MLP) has a good classification performance on rice HSIs because it removes translation invariance and local connectivity. Residual learning can improve the feature extraction ability of MLP network because of retaining the original information, preventing the model from degenerating, and facilitating the rapid convergence of the model. Therefore, a rice hyperspectral image classification model based on MLP network and residual learning is proposed. The results show that the proposed model has a higher classification accuracy (98.48%) than the other common classification models. In addition, the model has been verified on two public datasets with the accuracy higher than 99.95%.