A network based on deep learning for cloud detection in remote sensing, the paper is 《 Cloud Detection in Remote Sensing Images Based on Multiscale Features-Convolutional Neural Network[J]. 》
- loader_lc8.py
provides classe for loading landsat8-biome datasets. - saver_loader.py
provides the class and method of saving model and loading pre training model. - metrics.py
provides some evaluation indexes for model validation. - loss.py
provides the loss function used in the back propagation of training. There are two kinds of loss functions, namely "CE" and "focal". - Network.py
defines the architecture of the neural network, which is named MSNetwork. - train.py and valid.py
are the code of network training and validation respectively. The dataset used is 12 images in landsat8-biome, which is divided into 256 size patches, of which the training dataset accounts for 0.8, and the test dataset accounts for 0.2, and the data is scrambled during training.