The work aims to conduct non-destructive testing to the bacillus thuringiensis, beta-cypermethrin and indoxacarb on cauliflower with hyperspectral imaging technology, in order to detect pesticide residues on cauliflower quickly and non-destructively and track and study the safe interval for the best detection effects. After hyperspectral imaging was performed on cauliflower samples with and without pesticides, the spectral data of the region of interest were extracted. The original spectral data of the front and back 20 bands were eliminated to reduce the effect of noise, convolution smoothing method (S-G), multivariate scattering correction (MSC) and standard normal variate (SNV) algorithms were used to optimize the data of the remaining 216 bands (950-1 666 nm), respectively. The characteristic wavelength of spectral data of the three pesticides was extracted by the competitive adaptive reweighted algorithm (CARS) to improve the discriminant operation speed, and finally the partial least squares (PLS) discriminant model was established. The PLS model optimized based on SNV had the highest recognition accuracy of the three pesticides on cauliflower. The test results of the pesticide samples with indoxacarb were the best, with a recognition rate of 100%, and then the 7-day detection was carried out to this pesticide, of which the results were consistent with dissipation law of pesticides. The hyperspectral imaging technology has a good application prospect in the detection of pesticide residues on cauliflowers.