In order to realize the non-destructive testing of crop moisture content, taking lettuces of six water stress levels as experimental objects, the canopy moisture content of lettuce was detected and studied by using hyperspectral imaging technology and characteristic band selection method. Firstly, by analyzing the spectral reflectance of the canopy leaves and the background area, there were significant differences in spectral reflectance at 810.0nm and 710.7nm wavelengths, respectively. Therefore, the images of these two wavelengths were used to construct the mask image, which was used to mask the original hyperspectral image to remove background information. Secondly, spectral normalization was used to correct the light intensity of lettuce canopy. Thirdly, the standard normal variable (SNV) was used to preprocess the original spectral curve to eliminate the influence of scattering caused by particles on the sample surface. Fourthly, the irrelevant information was eliminated by Monte Carlo uninformative variable elimination (MCUVE), and then the least absolute shrinkage and selection operator (LASSO), successive projections algorithm (SPA), the least absolute shrinkage and selection operator coupled with successive projections algorithm (LASSO-SPA) were used to extract the characteristic wavelengths for data dimensionality reduction. Combing partial least squares (PLS), five lettuce canopy moisture content detection models were established. The results showed that the PLS model established by the full spectrum had the highest complexity and the worst predictive ability, because there were many redundant information variables and irrelevant variables in the full spectrum. The effect of PLS model with input variables screened by MCUVE-LASSO-SPA was the best. At this time, the correlation coefficients（R） of the modeling set and prediction set were 0.8827 and 0.9015, and the root mean square error （RMSE） were 1.0662 and 0.9287, respectively. The MCUVE-LASSO-SPA-PLS model was selected to calculate the dry basis moisture content of each pixel of the lettuce canopy, and a visual distribution map was generated to realize the visual detection of the dry basis moisture content of the lettuce canopy leaves. The research results provided a reference for the rapid non-destructive detection of lettuce canopy moisture content.