Accurate and efficient acquisition of organic matter content in different types of soil is of great significance to promote the prevention and control of soil degradation and the improvement of cultivated land quality in Northeast China. Totally 118 soil samples were collected from dryland farmland in Northeast China, including black soil, chernozem, fluvo-aquic soil and brown earth. The soil spectral information was obtained by ASD FieldSpec 4 spectrometer (350~2500nm). Reciprocal logarithm, first-order differential, continuum removal and continuous wavelet transform were used to preprocess the spectral curves. The relationship between the soil spectral and soil organic matter content was discussed. The optimal variable quantum set was screened by sCARS algorithm, and the partial least squares regression model was established. The results showed that continuous wavelet transform can not only effectively suppress the interference of background and noise, but also can excavate the effective information hidden in the soil spectrum, which greatly improved the correlation between the soil spectrum and organic matter content. Through the sCARS algorithm, redundant and overlapping spectral information variables were effectively removed, and important characteristic information variables related to soil organic matter were extracted, the efficiency of modeling was improved. The best models of black soil, chernozem, fluvo-aquic soil and brown earth were continuous wavelet transform model, with R2 reached 0.83, 0.88, 0.93 and 0.93, respectively. The first-order differential model also had good performance, but the modeling effect of reciprocal logarithm and continuum removal was not good. After continuous wavelet transform, the accuracy and stability of the soil organic matter hyperspectral inversion model were significantly improved. The R2 of the modeling set and validation set was increased by 0.13 and 0.28, and the RMSE was reduced by 2.48g/kg and 2.40g/kg, respectively. The continuous wavelet transform combined with the sCARS algorithm provided a way for hyperspectral prediction of soil organic matter, which can realize the rapid and accurate estimation of soil organic matter content.