Hyperspectral image classification method based on convolutional neural network

Abstract

The invention discloses a hyperspectral image classification method based on a convolutional neural network. The convolutional neural network in the method comprises an input layer, a spatial information extraction layer, a spectral information extraction layer, a full connection layer module and a classifier which are connected in sequence. The input layer is used for converting hyperspectral image pixels into input image blocks; the spatial information extraction layer is mainly used for extracting spatial information of the hyperspectral image; the spectral information extraction layer is mainly used for extracting spectral information of the hyperspectral image; the full connection layer module is used for converting the high-dimensional feature vector into a low-dimensional feature vector; and the classifier is used for normalizing the feature vectors obtained by the full connection layer module so as to predict and classify the category to which the hyperspectral image belongs. By applying the method, the spatial information and the spectral information of the hyperspectral image can be extracted at the same time, and the classification speed and the classification precisioncan be well balanced.

Description

technical field [0001] The invention relates to the technical field of image classification, in particular to a convolutional neural network-based hyperspectral image classification method. Background technique [0002] Hyperspectral imaging technology uses dozens to hundreds of continuous and subdivided spectral bands to simultaneously image the target area. While obtaining surface image information, it also obtains its spectral information. For the first time, it has truly achieved the combination of spectrum and image. Hyperspectral remote sensing has been widely used in the field of earth observation in recent years. Efficient hyperspectral classification technology will greatly improve the classification accuracy of ground objects, thereby improving the efficiency of earth observation. [0003] Convolutional Neural Network (CNN) is one of the mainstream technologies in the field of hyperspectral image classification. People hope to have a CNN model that can meet the req...

AI Technical Summary

Problems solved by technology

The deep learning model proposed by Lee et al. is small, which is convenient for quickly completing the hyperspectral image classification task, but due to the limitations of the model, it cannot extract enough spatial information, so the accuracy of the model is limited. For example, in the public dataset University of In Pivia and Indian Pines, under the condition of 200 training samples for each type, the average accuracy of the test set is 95.79% and 93.61%, respectively

In order to extract enough spatial information of hyperspectral images, Zhang et al. proposed a DRCNN model (Zhang, Mengmeng, Li, Wei, Du, Qian.Diverse Region-Based CNN for Hyperspectral Image Classification[J].IEEE Transactions on Image Processing, 2008:1-1.), which has many parallel convolution modules and significantly improves the test accuracy, but the DRCNN model needs to consume a lot of computing resources, and the training and testing process takes too much time. Its data set University The test time of Pivia is 5.52 times that of the aforementioned Lee model

Paoletti et al. proposed a convolutional neural model (Paoletti M E, Haut J M, Plaza J, et al. A new deep convolutional neural network for fast hyperspectral image classification [J]. Isprs Journal of Photogrammetry&Remote Sensing, 2018, 145PA (NOV.) :120-147.) is used to quickly complete hyperspectral image classification, and its classification accuracy is between Lee model and Zhang model. In fact, although Paoletti's model reduces the model depth and has no branches, its input and convolutional layer filtering The size of the filter is significantly larger than the previous two methods. For example, in the dataset University of Pivia, the optimal input size of the Paoletti model is 21×21×the number of channels, while the best inputs of the Lee model and Zhang model are 5×5×channels. number and 9×9×number of channels, which leads to a significant increase in the test time of the Paoletti model

[0004] The above methods either focus on classification speed or classification accuracy, or reduce the depth and complexity of the model in order to balance classification accuracy and speed, but the classification speed is still not satisfactory

Images