Egypt’s Remote Sensing Land Use Classification Using Deep Learning

Egypt’s landscape is changing rapidly and continuously and this is highly affecting the developing of the economy. No tools other than remote sensing for monitoring the land use is effective. Millions of land images are widely available today from the satellite images, these land images have to be classified into different categories to ensure proper land management and land decision making. Our main objective is to explore the use of deep learning for the land classification of Egypt’s remote sensing images. Our chapter exploits different deep convolutional neural network models to extract features from the images followed by category classification by supervised classifiers. We report at the end the accuracy and the performance comparisons between the different testing models on three different standard datasets and on Egypt’s land images. Standard datasets are used to fine-tune the classifier layers of the pre-trained CNN networks AlexNet, VGG16 and ResNet. In general the SVM classifier outperforms the other two tested classifiers (KNN and the Naïve Bayes). The highest accuracy, 94.7%, achieved by ResNet model on the RS19 dataset. We obtained our dataset from Google Earth Engine from different parts of Egypt. The Egyptian dataset is used finally for testing without retraining the classifier layers to test the ability of the models for real-time applications. We achieved a highest accuracy of 60% with AlexNet.