Optical remote sensing images contain complex texture features. The noise in remote sensing images affects not only the visual effect of images but also the processing, analysis, transmission, and storage of images. Therefore, image denoising becomes an important step in remote sensing image processing. Traditional denoising methods are likely to cause problems such as loss of image details and blurred denoising results. Recently, deep learning has been rapidly developing in the field of image denoising, compared with traditional algorithms, the stability of the denoising algorithm of deep learning algorithms has improved tremendously. However, the real noise in remote sensing images and the reconstruction of the image after denoising, is the main problem in the field of image denoising at present. In this paper, an MRFENet remote sensing image denoising algorithm based on multi-sensory field feature fusion and enhancement is proposed. To address the problem that image details are lost after denoising and real noise is difficult to be eliminated, the following approach was used. First, a global feature extraction module is introduced, which consists of several convolutions with different dilation rates, followed by fusion of the extracted features. The purpose of this process is to allow the model to expand the receptive field without increasing the number of parameters, and to enable the model to converge quickly by extracting shallow features at different scales. Second, multi-scale feature enhancement blocks are introduced. Each block consists of a multi-scale feature extraction layer and a channel attention module, both of which form the residual structure. The purpose is to be able to extract multi-scale features at different levels and to assign higher weights to important features to achieve enhancement of important features. The residual structure ensures that the network does not explode in gradient due to excessive depth. Finally, in order to reduce the loss of feature information and the fluctuation caused by the fusion of shallow features with deep features, the resulting features at different levels are chosen to be fused step by step to enhance the continuity of pixels. To make the denoised images more consistent with the visual perception, MS-SSIM is chosen as the loss function during the training process. The number of channels and the number of multi-scale feature enhancement blocks of the MRFENet are configurable, and the performance of the network does not increase with the number of modules, so the most suitable network parameters can be obtained by combining the network performance with the computational effort. In order to test the denoising ability of MRFENet for remote sensing images of different sizes, two publicly available remote sensing image datasets with different sizes are selected. By adding different intensity of noise on each dataset, this paper tests the denoising stability of MRFENet for different intensity of noise. In order to test the denoising performance of MRFENet for real noise, a hyperspectral real remote sensing image is selected for testing. PSNR and SSIM are selected as quantitative evaluation metrics for different intensity noise datasets to evaluate the denoising results. NIQE, BRISQUE, PIQE are selected as quantitative no-reference evaluation metrics for real noise datasets to evaluate the denoising results. After comparing the denoising results with those of traditional denoising algorithms NLM, BM3D and deep learning algorithms DnCNN, RIDNet and REDJ, it can be concluded that the proposed algorithm has the best performance on each dataset and outperforms other algorithms in all metrics. The images denoised with MRFENet can retain the edge details and do not show excessive smoothing, which is more in line with the visual perception. The effectiveness and generalization of MRFENet algorithm for remote sensing image denoising are verified.