/Hyperspctral_classification

Hyperspectral remote sensing image classification based on deep learning

Primary LanguagePython

a multi-scale 3D convolutional neural network is used in hyperspectral image classification.

Overview:

this is my graduation project that classify indian_pines、Pavia_university and Salinas dataset based on deep_learning. download the dataset

Language:

Python3.x

Tools:

Tensorflow

Describe：

The sensor source of image does not require much pre-processing, in order to reduce the burden on the computer, the input data is compressed using PCA and 96% effective spectral information is extracted. The data is normalized and input into a three-dimensional convolutional network to get feature of empty and spectrum. The three-dimensional convolutional network uses multi-scale convolution kernels to extract multi-scale spatial features, then the joint feature maps of the spectral and spatial properties of the hyperspectral image fed through a fully connected layer, which finally predicts each pixel label through the Softmax classifier.