Aiming at the need for automatic identification of haze concentration in the adaptive dehazing system, this paper presents an algorithm for haze image classification based on Support Vector Machine (SVM) and mixed feature. On the basis of the characteristics of the haze image, the mixed feature vector composed of the dark channel feature, the wavelet feature and the Mean Subtracted Contrast Normalized (MSCN) feature is adopted to describe the characteristic differences of the images with different haze concentration. The SVM classifier implements the automatic identification and classification of haze images through supervised learning of the mixed feature vectors. Experimental results show that the method can effectively identify the images of haze-free, thin haze and dense haze, which provides a good basis for the dehazing system to select dehazing parameters adaptively based on the haze concentration.