Abstract:Spacecraft image denoising is a crucial basic technology closely related to aerospace research. However, the existing deep learning-based image denoising methods lack deep consideration of the characteristics of spacecraft image. To address the aforementioned shortcomings, we analyses spacecraft noise image and identifies two main characteristics. One is that there are a large number of low-light images in the obtained spacecraft noise image dataset. Another is there are a lot of repetitive periodic features in spacecraft image. According to the above mentioned characteristics, we propose a Edge modeling Activation Free Fourier Network (EAFFN), which is an efficient spacecraft image denoising method including Edge Modeling Block (EMB) and Activation Free Fourier Block (AFFB). We present EMB to effectively model edge and extract structural information and better identify the spacecraft components from dark regions in spacecraft noise image. We present AFFB and utilize an improved fast fourier block to extract repetitive periodic features and long-range information in noisy spacecraft image. In addition, Simple Gate is designed in our AFFB to reduce the computational complexity. Extensive experimental results demonstrate our EAFFN performs competitively to the state-of-the-art on spacecraft noise image datasets.

What problem does this paper attempt to address?

This paper attempts to solve the problem of spacecraft image denoising. Specifically: 1. **Deficiencies of existing methods**: Existing deep - learning - based image denoising methods lack in - depth consideration of the characteristics of spacecraft images. These methods fail to fully utilize emerging deep - learning techniques or are not designed according to the characteristics of spacecraft noisy images. 2. **Characteristics of spacecraft images**: - **Low - light images**: Due to the harsh lighting conditions in the space environment, many acquired spacecraft images exhibit low - light characteristics. - **Repeated periodic structures**: As man - made objects, spacecrafts have relatively regular overall structures and a large number of repeated periodic components (such as solar panels). 3. **Proposed method**: To solve the above problems, the authors propose the Edge Modeling Activation Free Fourier Network (EAFFN), which is an efficient spacecraft image denoising method, including the Edge Modeling Block (EMB) and the Activation Free Fourier Block (AFFB). 4. **Specific contributions**: - **EMB**: Effectively extract edge and structure information in low - light spacecraft images through trainable Sobel convolutions, enhancing the ability to distinguish spacecraft components in noisy images. - **AFFB**: Utilize the improved fast Fourier transform block to extract repeated periodic features and long - distance global information, and design the Simple Gate to reduce computational complexity. Through extensive experimental results, EAFFN performs excellently on the spacecraft noisy image dataset, significantly outperforming existing methods and having a lower computational complexity. ### Formula summary - **EMB process**: \[ X_{\text{edge}}=\text{DSConv}(\text{SConv}(I_{\text{noisy}})) \] where \(I_{\text{noisy}}\) is the input noisy spacecraft image, \(X_{\text{edge}}\) is the output enhanced edge feature, and \(\text{SConv}\) and \(\text{DSConv}\) represent 3×3 Sobel convolution and 3×3 depth Sobel convolution respectively. - **AFFB process**: \[ X_{\text{SG1}}=\text{SG}(\text{DConv}(\text{Conv}(\text{LN}(X_{\text{in}})))) \] \[ X_{\text{SFFB}}=\text{SFFB}(X_{\text{SG1}}) \] \[ X_{\text{F}}=\text{Conv}(X_{\text{SFFB}})+\alpha * X_{\text{in}} \] \[ X_{\text{SG2}}=\text{SG}(\text{Conv}(\text{LN}(X_{\text{F}}))) \] \[ X_{\text{AFFB}}=\text{Conv}(X_{\text{SG2}})+\beta * X_{\text{F}} \] where \(X_{\text{in}}\) and \(X_{\text{AFFB}}\) are the input and output feature maps of AFFB respectively, \(\text{LN}\) represents layer normalization, \(\text{Conv}\) and \(\text{DConv}\) represent the convolution layer and the depth convolution layer respectively, \(\text{SG}\) represents the Simple Gate, \(\text{SFFB}\) represents the simplified fast Fourier block, and \(\alpha\) and \(\beta\) are trainable parameters that control the weights of skip connections. - **SFF

Edge Modeling Activation Free Fourier Network for Spacecraft Image Denoising

Spatial-Adaptive Network for Single Image Denoising

Frequency-Domain Deep Guided Image Denoising

Multitask Sparse Representation Model-Inspired Network for Hyperspectral Image Denoising

Learning a Model-Based Deep Hyperspectral Denoiser from a Single Noisy Hyperspectral Image.

Enhanced Frequency Fusion Network with Dynamic Hash Attention for image denoising

FEUSNet: Fourier Embedded U-Shaped Network for Image Denoising

Edge Prior Augmented Networks for Motion Deblurring on Naturally Blurry Images

An efficient lightweight network for image denoising using progressive residual and convolutional attention feature fusion

The LiDAR Echo Denoising Algorithm Based on EMD and FFT

Research Advanced in Image Denoising Based on Deep Learning

Multi-Scale Deep Neural Network Based on Dilated Convolution for Spacecraft Image Segmentation

DFANet: Denoising Frequency Attention Network for Building Footprint Extraction in Very-High-Resolution Remote Sensing Images

Removal of multisource noise in airborne electromagnetic data based on deep learning

ANED-Net: Adaptive Noise Estimation and Despeckling Network for SAR Image

CFPNet: a denoising network for complex frequency band signal processing

Frequency Information Enhanced Deep EEG Denoising Network for Ocular Artifact Removal

Removing multiple types of noise of distributed acoustic sensing seismic data using attention-guided denoising convolutional neural network

Suppression of seismic random noise by deep learning combined with stationary wavelet packet transform

A Study on Denoising with Deep Convolutional Neural Networks in Spatial Heterodyne Spectroscopy

Remote Sensing Image Denoising Based on Deep and Shallow Feature Fusion and Attention Mechanism