Evaluate the effect of changes in the Cloud-Net+ model's training parameters for cloud detection in Landsat 8 Remote Sensing Imagery

  • Safeyya A. Alshehhi

Student thesis: Master's Thesis


Cloud detection is one of the basic stages to process optical remote sensing image analysis. There are multiple techniques to detect the existence of the clouds. Those techniques are mostly taking the cloud's physical or optical properties as parameters. Some machine/deep learning approach employed for cloud detection in satellite imagery include but not limited to, multi-scale cloud net (MSCN), clear-sky background differentiation (CSBD), support vector machine (SVM), K-nearest neighbor (KNN), fully convolutional neural network (FCN) and convolutional neural network (CNN). Of all these machine/deep learning approaches, CNN has proven to be very robust for cloud detection in many satellite images, under various atmospheric conditions and in different geographic locations. The objective of this research is to evaluate the effect of changes in the training parameters of CNN-based model for cloud detection (Cloud-Net+) on Landsat 8 Remote Sensing Imagery. This model is using a unique Filtered Jaccard loss function for the CNN model. Choosing the right parameters for training is essential for having a model that is suitable for many datasets where it's not overfitting or underfitting a specific dataset. The decision of selecting those parameters depends on the use of the model. We will be we presenting the results of changing a selected training parameter and compare them with the original results. The accuracy of the model was enhanced with the new parameters. The highest accuracy reached was 97.22% using the optimum parameters compared with the original accuracy which was 96.36%. On the other hand, the Cloud-Net+ model outperformed some of the others cloud detection models with both the original values and the enhanced training parameters.
Date of AwardJul 2020
Original languageAmerican English


  • CNN
  • machine learning
  • cloud detection
  • satellite images.

Cite this