Reconstructive micro-spectrometers have shown great potential in many fields such as medicine, agriculture, and astronomy. However, the performance of these spectrometers is seriously limited by the spectral varieties of response pixels and anti-noise ability of reconstruction algorithms. In this work, we propose a spectral reconstruction (SR) algorithm, whose anti-noise ability is at least four times better than the current algorithms. A micro-spectrometer is realized by fabricating a large number of Fabry-Perot (FP) micro-filters onto a cheap complementary metal-oxide semiconductor (CMOS) chip for demonstration by using a very high-efficiency technology of nano-imprinting. Nano-imprint technology can complete hundreds of spectral pixels with rich spectral features at one time and with low cost. In cooperation with the SR algorithm, such a micro-spectrometer can have a spectral resolution as high as 3 nm with much lower angular sensitivity than a photonic crystal-based micro-spectrometer. It can obtain the target's spectrum from only a single shot, which has wide applications in spectral analysis etc.