【摘要】 目前，像素内量子效率的非线性特性尚缺乏系统性的研究与分析，且非线性与线性标校方案的成像效果对比仍属空白。现有的成熟技术方案严格按照线性来处理像素量子效率，忽视了像素表面材料的不均匀性和非线性，增加了成像偏差，这种偏差在高要求的成像中不可忽视。文中使用稳态低频动态正弦干涉光场和时域拟合算法，求解了亚像素非线性量子效率，并应用于高质量超分辨成像中，验证了非线性标校方案的优越性。实验表明，相较于线性标校，基于非线性量子效率的超采样成像可使系统调制传递函数(Modulation Transfer Function, MTF)提升30%以上。更多还原

【Abstract】 Objective The precision of digital image sensors(DIS) is fundamentally limited by the assumption of linear pixel response in conventional quantum efficiency(QE) calibration. This assumption fails to account for the inherent non-uniformity and nonlinearity of photoelectric conversion at the sub-pixel level, leading to imaging deviations that are particularly critical in high-precision applications like remote sensing and astronomy. While intra-pixel QE has been recognized as a key factor, systematic research into its nonlinear characteristics and a comparative analysis of imaging performance between nonlinear and linear calibration schemes have been absent.This study aims to address this gap by establishing a nonlinear sub-pixel QE model, developing a novel calibration method using a steady wave field, and demonstrating its superior performance in hyper-sampling imaging(HSI).Methods A nonlinear model for sub-pixel QE was developed by conceptually dividing a pixel into k×k subregions and representing the photoelectron-photon relationship with a second-order Taylor expansion(Eq.(1)).The coefficients of this model were determined by illuminating the sensor with a steady, low-frequency, dynamic sinusoidal interference pattern generated by two laser beams. The precise expression of this input wave field was fitted in the time domain(Eq.3). By capturing a series of images of this moving pattern and solving the system of equations(Eq.5), the intra-pixel nonlinear QE(NLQE) distribution was accurately measured(Fig.2). This premeasured NLQE was then applied in a recomputation process to reconstruct high-resolution images from a sequence of k2 images of the target captured with relative displacements(Fig.3), a technique we term Nonlinear Hyper-Sampling Imaging(NLHSI).Results and Discussions The NLQE of a Hamamatsu C14041-10U near-infrared CCD was measured and compared to its linear counterpart(LQE). Results showed that the NLQE varied significantly with illumination intensity, especially under very low and near-saturation conditions, while the LQE remained constant(Fig.4(a)-(d)). This confirms that NLQE more accurately captures the complex physical response of the sensor. When applied to HSI, the NLHSI method demonstrated a remarkable improvement in image quality. At the sensor’s Nyquist frequency, the modulation transfer function(MTF) of NLHSI showed an improvement of ≥150%compared to the original direct imaging, and a further ≥30% improvement over HSI using linear QE calibration(Linear NHSI, or LNHSI)(Fig.4(e)). Practical imaging tests, including near-field printed text and a building 6 km away, further validated these findings. The text recognition accuracy for images reconstructed using NLHSI was significantly higher than that for LNHSI and the original images(Fig.5). Furthermore, compared to deep learningbased image super-resolution schemes, NLHSI exhibits a clear advantage in stability.Conclusions This work successfully establishes and validates a framework for nonlinear sub-pixel quantum efficiency calibration and its application in hyper-sampling imaging. The proposed NLHSI technique physically enhances the sampling resolution and image quality of conventional DIS without any hardware modifications. By providing a more accurate characterization of the sensor’s response, NLHSI achieves substantial gains in MTF and information retrieval capability over both native imaging and linearly-calibrated HSI. This software-based enhancement strategy offers a low-cost, high-impact pathway to significantly boost the performance of imaging systems, with immediate potential for application in aerospace remote sensing, security surveillance, and lowaltitude monitoring, particularly in the infrared and near-infrared bands.更多还原