This research focused on the detection of trashes at the depth of 1～6mm in the cotton using hyper-spectral imaging and pixels classification. In the wavelength range of 460～900nm, the detection algorithms were developed based on pixel’s spectra of hyper-spectral image, pixel classification by quadratic discriminate analysis, and binary images post processing by combining area filter with morphologic process. The results indicated that hyper-spectral imaging was able to detect some trashes at certain depths in the cotton, such as natural trashes, color polypropylene fibers, color yarn and fragments of cloth. And it also can detect some black hair and gray polypropylene fibers. In particular, the detection effect for natural trashes was the best. The detection rates for natural trashes were over 80%.