Now pulse signal feature is difficult to be extracted in the sub-health detecti on,and it is easy to affect recognition accuracy due to the relying on mainly hand-crafted feature e xtraction.To solve these problems,this paper proposes a pulse signal sub-health detection method based o n principal component analysis network (PCANet).Firstly,PCANet is used to extract features from prep rocessing pulse signal in the experiment.Then,we deal these features with Hash code,histogram block as the description features.Finally,the feature vectors are fed into two classifiers to classify the healthy and sub-healthy subjects.The test results is compared with the results from other methods of traditional features extraction.The experimental analysis show that our method obtains the highest accuracy rate in sub-health recognition field.The recognition rate of PCANet method is improveed by 10% compared with those of traditional feature extraction methods.So our approach can effectively distinguish health and sub-health subjects,and provides a new reference for human detection of sub -health state.