The vegetation coverage is one of the most effective indicators to evaluate the desertification of land, and also is an important indicator for monitoring vegetation. Hyperspectral estimation of vegetation coverage can provide an important basis for vegetation monitoring. Tribulus terrestris L. fixed sand in the semi arid zone Togtoh County was as the research object, the spectral curve features of different vegetation coverages were analyzed; the correlation between the original spectral vegetation index of two wave bands and the vegetation coverage was extracted, and the optimal wave band combination was selected. The spectral reflectance ratio of vegetation was decomposed at the different scales by the continuous wavelet transform (CWT), and the optimal wave bands at the different decomposition scales were extracted. The vegetation coverage estimation models were established using the partial least square method (PLSR) and support vector machine (SVM) with the different independent variables. The results showed that: ① there was significant correlation between the original spectrum vegetation index and vegetation coverage with all the correlation coefficients above 0.55, and the optimal band combinations were DI (2 260 and 2 210 nm), RI (1 410 and 660 nm), NI (1 470 and 670 nm)，RDVI(1 770 and 670 nm) and MSR(1 410 and 660 nm), respectively. ② There was also good correlation between the wavelet coefficient and vegetation coverage, which the correlation coefficients in the 1~10 scales of original spectral were high than 0.72 with the maximum correlation 0.788 9 in the sixth decomposition scale at the band 630 nm. The maximum correlation between the wavelet coefficient extracted from the first-order differential spectrum and the vegetation coverage was 0.806 9, and the correlations between the wavelet coefficient of 1~10 scale extracted from the second-order differential spectrum and the vegetation coverage were above 0.6 with the maximum correlation 0.781 8. ③Among the models established by using the original spectral vegetation index and wavelet coefficient extracted by CWT as the independent variables, the PLSR model input the second-order differential wavelet coefficient showed best precision and stability, which R² was 0.905 9 and RMSE was 0.035 6, respectively. These results showed that the CWT algorithm could improve spectral characteristic information, which provided a technical method for the estimated inversion of vegetation coverage.