This paper proposes a novel deep learning-based method for remote sensing image enhancement called GSA-Net, which integrates multiscale feature fusion and global spatial attention mechanisms. The method aims to enhance brightness while preserving image details. To address the challenge of limited training data, gamma correction is applied to generate low-light images, which are used as training examples. A loss function combining Structural Similarity (SSIM) and Peak Signal-to-Noise Ratio (PSNR) is introduced to guide the model's training. The GSA-Net network is tested on the NWPU VHR-10 dataset, demonstrating superior performance compared to state-of-the-art algorithms in terms of PSNR, SSIM, and Learned Perceptual Image Patch Similarity (LPIPS). Additionally, the method provides better remote sensing images with distinct details and higher contrast for high-level visual tasks like object detection. The GSA-Net model employs a U-shaped network with multiscale sampling and a global spatial attention (GSA) module to enable feature interaction across channels, suppressing redundant information and improving low-scale texture details and multilevel features. The model also incorporates depthwise separable convolution to reduce parameters and computations, and a selective kernel feature fusion (SKFF) module to effectively integrate features during reconstruction. The proposed loss function combines PSNR and SSIM to avoid model optimization direction deviation and gradient diffusion, enhancing convergence and performance. The method is evaluated on the NWPU VHR-10 dataset, with results showing that GSA-Net achieves state-of-the-art performance in image enhancement and facilitates object detection on enhanced images. The model's lightweight design and efficient computation make it suitable for practical applications. The study also compares the proposed method with other algorithms, demonstrating its effectiveness in enhancing low-light remote sensing images and improving visual quality. The results indicate that GSA-Net is capable of learning features that conform to visual patterns, making it a promising approach for remote sensing image enhancement.This paper proposes a novel deep learning-based method for remote sensing image enhancement called GSA-Net, which integrates multiscale feature fusion and global spatial attention mechanisms. The method aims to enhance brightness while preserving image details. To address the challenge of limited training data, gamma correction is applied to generate low-light images, which are used as training examples. A loss function combining Structural Similarity (SSIM) and Peak Signal-to-Noise Ratio (PSNR) is introduced to guide the model's training. The GSA-Net network is tested on the NWPU VHR-10 dataset, demonstrating superior performance compared to state-of-the-art algorithms in terms of PSNR, SSIM, and Learned Perceptual Image Patch Similarity (LPIPS). Additionally, the method provides better remote sensing images with distinct details and higher contrast for high-level visual tasks like object detection. The GSA-Net model employs a U-shaped network with multiscale sampling and a global spatial attention (GSA) module to enable feature interaction across channels, suppressing redundant information and improving low-scale texture details and multilevel features. The model also incorporates depthwise separable convolution to reduce parameters and computations, and a selective kernel feature fusion (SKFF) module to effectively integrate features during reconstruction. The proposed loss function combines PSNR and SSIM to avoid model optimization direction deviation and gradient diffusion, enhancing convergence and performance. The method is evaluated on the NWPU VHR-10 dataset, with results showing that GSA-Net achieves state-of-the-art performance in image enhancement and facilitates object detection on enhanced images. The model's lightweight design and efficient computation make it suitable for practical applications. The study also compares the proposed method with other algorithms, demonstrating its effectiveness in enhancing low-light remote sensing images and improving visual quality. The results indicate that GSA-Net is capable of learning features that conform to visual patterns, making it a promising approach for remote sensing image enhancement.