Objective Diffraction optical systems are a core focus for large-aperture space optical payloads due to their integration, lightweight, and ultra-large aperture—advantages that overcome the limitations of traditional catadioptric systems (low degrees of freedom, bulky structure). However, they suffer from notable stray light-induced image degradation: non-design-order diffracted rays cause low resolution, poor contrast, and dark-field detail loss. Traditional enhancement methods (e.g., deconvolution) rely on accurate Point Spread Function (PSF) modeling, leading to on-orbit model mismatch; deep learning is hindered by high power/latency, while CPU/GPU fail to meet on-orbit real-time/low-power needs. Fixed-parameter gamma correction also cannot adapt to dynamic environments. This study aims to design an FPGA-based low-latency, low-power hardware architecture for on-orbit real-time diffraction image enhancement.

Methods An integrated enhancement algorithm framework is proposed: 1) Information entropy-driven automatic gamma correction: By calculating the information entropy of images under gamma parameters ranging from 0.1 to 1.5 and selecting the optimal γ, this algorithm expands dark-field details suppressed by stray light and improves overall image brightness. 2) Unsharpen Mask (USM): Adopting USM to solve the edge detail blurring caused by stray light. The core of USM lies in Gaussian filtering to extract high-frequency edge details: A 5×5-sized Gaussian convolution kernel is selected, and a daisy-chain FIFO row cache structure is used to realize the sliding window operation of Gaussian filtering (Fig.6). To improve real-time performance and throughput, a four-level pipeline operation architecture is designed (Fig.7). 3) Automatic image adjustment module: Extreme noise is filtered via histogram quantile thresholds, grayscale data is linearly stretched to the range 0, 255, and secondary gamma correction is triggered if the grayscale median (Mt) is less than 128.

Results and Discussions For 640×512 8-bit diffraction images, the proposed FPGA architecture achieves a processing latency of 13 ms and power consumption of 1.464 W which means that the running speed of the architecture is 38 times higher than that of the CPU, and the power consumption is less than 1/5 (Tab.8). Key image quality metrics are significantly improved: For a typical test image (Image ②), the mean gradient (MG) after "Gamma+USM" processing is 31% higher than that of single Gamma correction; the final PSNR (14.22 dB) exceeds Gamma correction alone (12.24 dB), while structural similarity (SSIM) is well-maintained to avoid excessive distortion. The FPGA resource occupancy is moderate, fully meeting the resource constraints of on-orbit satellite payloads.

Conclusions The FPGA-based hardware architecture effectively reduce stray light-induced image degradation via the integrated optimized algorithm framework and parallel/pipeline hardware design. It fully satisfies the dual technical requirements of on-orbit satellite systems for low latency (13 ms) and low power consumption (<1.5 W), providing a reliable solution for on-orbit real-time processing of diffraction imaging systems.