【摘要】 单像素成像作为一种新型的计算成像技术，在获取高质量图像时需要多次测量，严重影响了成像速度。为实现高效成像，提出一种基于阵列空间光场的端到端傅里叶单像素成像方法。该方法基于双残差网络，通过引入全局注意力机制，显著提升了网络性能。利用2×2阵列傅里叶空间光场调制获取一维探测值信号，并将探测值及其对应的目标图像标签输入到优化的双残差网络中，经过训练后能够有效重构出高质量的目标物体图像。数值仿真结果表明，相同采样率下，该方法的成像质量优于传统的随机散斑、自然序Hadamard和傅里叶单像素成像方法的，即使在采样率低至0.195%时，仍然能够重构出清晰的目标图像。更多还原

【Abstract】 Single-pixel imaging, as a new type of computational imaging technology, requires multiple measurements to obtain high-quality images, which significantly affects the imaging speed.To achieve efficient imaging, an end-to-end Fourier single-pixel imaging method based on array spatial light field is proposed.The method is based on a dual residual network and significantly improves the network performance by introducing a global attention mechanism.Using a 2×2 array Fourier spatial light field modulation to obtain one-dimensional detection value signals, and inputting the detection values and their corresponding target image labels into the optimized dual residual network, after training, it can effectively reconstruct high-quality target object images.Numerical simulation results show that, under the same sampling rate, the imaging quality of this method is superior to traditional random speckle, natural order Hadamard, and Fourier single-pixel imaging methods, even when the sampling rate is as low as 0.195%,it can still reconstruct clear target images.更多还原