Illegal tree cutting activities such as clearing forests inside protected areas for personal needs are contributing high to environmental degradations. Though it is a big factor to climate change, reduced biodiversity, habitat loss and other devastating effects, it is roughly monitored or not monitored at all. Thus, solutions need to be developed to detect and handle tree cuttings. With advances in remote sensing data, deforestation detection has become more convenient and accurate. Researches have been done on deforestation detection from satellite images but the methods and the satellite data used so far are not appropriate for study areas with possibilities of finding heterogenous land cover classes within small areas. The coarser spatial resolution of the satellite images, the usage of traditional statistical methods as classifiers, and the difficulty in optimal patch size selection when patch-based classification is used were the identified problems of the previous researches. As a solution to these problems, we proposed a deep learning based semantic segmentation model to be used as a land cover classification scheme at pixel-level before tree cutting detection is taken place. In the proposed model, high resolution sentinel-2 satellite images of our study area (Gambella National Park) were used as a dataset, at pixel-level classification was performed and deep learning architectures were employed. We have built the U-Net architecture first and its encoder part was modified further to incorporate the state-of-the-art Convolutional Neural Network (CNN) architectures such as Vgg-16 and ResNet-50. The U-Net model and the U-Net model with modified encoders were ensembled to build the final semantic segmentation model. The original U-Net model, U-Net model with Vgg-16 encoder and U-Net model with ResNet-50 encoder achieve 86.69%, 88.08%, 91.28% average F1-Scores respectively. And the final ensembled model achieves 92.12% average F1-Score. Finally, we have used the post-classification comparison approach to perform tree cutting detection in our study area. Sentinel-2 images of our study area captured in 2021/11/30 and 2021/12/10 were automatically downloaded, pre-processed and classified into land cover classes using the ensembled model. The classification outputs were then compared to detect and map the tree cuttings in our area of interest. Keywords: Satellite Images, U-Net, Semantic Segmentation, Tree Cutting, Gambella National Park