This paper introduces a novel pre-training framework called Generic Knowledge Boosted Remote Sensing Pre-training (GeRSP) to enhance the performance of deep learning models in remote sensing image understanding tasks. GeRSP addresses the domain gap between remote sensing images and natural images by combining supervised and self-supervised pre-training methods. The framework includes two branches: a self-supervised branch for learning domain-specific features from unlabeled remote sensing images, and a supervised branch for learning general knowledge from labeled natural images. These branches are integrated using a teacher-student architecture to simultaneously learn representations with both general and domain-specific knowledge. The effectiveness of GeRSP is evaluated on three downstream tasks: object detection, semantic segmentation, and scene classification. Experimental results demonstrate that GeRSP consistently improves the performance of these tasks compared to other pre-training methods, highlighting its ability to learn robust representations and bridge the domain gap. The paper also includes visual analysis and sensitivity analysis to further evaluate the effectiveness of GeRSP.This paper introduces a novel pre-training framework called Generic Knowledge Boosted Remote Sensing Pre-training (GeRSP) to enhance the performance of deep learning models in remote sensing image understanding tasks. GeRSP addresses the domain gap between remote sensing images and natural images by combining supervised and self-supervised pre-training methods. The framework includes two branches: a self-supervised branch for learning domain-specific features from unlabeled remote sensing images, and a supervised branch for learning general knowledge from labeled natural images. These branches are integrated using a teacher-student architecture to simultaneously learn representations with both general and domain-specific knowledge. The effectiveness of GeRSP is evaluated on three downstream tasks: object detection, semantic segmentation, and scene classification. Experimental results demonstrate that GeRSP consistently improves the performance of these tasks compared to other pre-training methods, highlighting its ability to learn robust representations and bridge the domain gap. The paper also includes visual analysis and sensitivity analysis to further evaluate the effectiveness of GeRSP.