The combination of broadband filtering arrays and compressed sensing algorithms is an effective solution for achieving miniaturized spectrometers. However, high correlation in the measurement matrix can worsen spectral reconstruction errors. To address this, a spectral measurement and reconstruction scheme combining simulated soil microspore structure and the difference method is proposed. First, by simulating the microscopic pores of the soil, a set of initial filtering units with lower correlation is constructed through a high-degree-of-freedom random internal pore pattern. Next, correlation analysis is conducted, and a difference method is adopted to weaken the impact of shared features among different filtering units on correlation. Finally, the 64 channels with the least correlation are selected to construct the measurement matrix, and the compressed sensing algorithm is used to reconstruct the spectrum to be measured. The simulation results show that, compared to traditional schemes, when the average area ratio of the filtering pattern is 0.4, the proposed method using the simulated soil micropore structure and difference method reduces the correlation from 0.2867 to 0.0992, and the reconstruction error for different bandwidth spectra decreases by over an order of magnitude. This design approach can provide valuable insights for high-precision intelligent sensing.