A method based on wavelet packet energy spectrum feature extraction and support vector machine recognition is presented in this paper to solve the problem of recognizing the floating and suspending impurities in ampoules. Firstly

the impurity zone’s image is obtained through image sequence difference and points detection division. Then

a 1-D signal can be obtained through adding the ROI row by row in axis direction of an ampoule. The 1-D signal was decomposed by wavelet packet

and the independent primary components in wavelet packet feature vector are extracted using PCA

and the wavelet packet energy spectrum of the independent primary components is taken as the feature of impurity types. As the input vector of support vector machine

the sample features can be classified rapidly by sequential minimal optimization method through training. Different types of core functions and corresponding parameters are selected for training and testing in the experiments. The results of experiments show that the recognition period of SVM decrease by 60% and the recognition precision is improve by 35%

as compared to the BP network. This method can meet the requirement on feature extraction and rapid recognition for the floating particles in production.