Keras implementation of land use classification by CNNs

• Simpilified Residential Network: ./Code/Simplified_ResNet.py
• VGG-like Network: ./Code/AtrousVGG.py
• Two-stream convolutional neural network for combining features (CFCNN): ./Code/Keras_Merge.py
• Classify test samples by single data-based model: ./Code/Classify_TestSample_SingledataCNN.py
• Classify test samples by multi-source data-based model: ./Code/Classify_TestSample_CFCNN.py
• Classify the whole research area by trained models: ./Code/Classify_FullImage.py

The source code is compiled on the Windows 10 platform using Python 3.6. The dependencies include:

tensorflow-gpu: 1.9, backend
Keras: 2.2.4, framework
pandas: used for csv I/O
numpy: used for array operations
matplotlib: used to visualize training accuracy curves
GDAL: used for remote sensing image I/O
scikit-learn: used for data preprocessing

We provide training data and test data for estimating the performance of models. Training data and test data can be found in ./Data, stored as sample points.

The original high spatial resolution image and population density data can be downloaded from Baidu Netdisk. The extracted code is fo7v.

We provide a sample of data in ./Data/Exam_ClassificationResult. The folder contains processed high-resolution images and population density data. We can test the feasibility of the code in ./Code/Classify_FullImage.py.