To solve some problems of the single image smoke detection, such as, the study of single image smoke detection is less and smoke detection method of single image based on color model is easily influenced by the color suspected smoke area, we propose a Multilevel filter algorithm of single image smoke detection. Firstly, for the dark channel transmittance chart optimized by the guide filter, the algorithm carries out binaryzation on the image according to certain threshold, which make it fast to remove the interference area of high transmittance and obtain the low transmittance area. Then it removes the smoke in the low transmittance area, extracts the Cr channel of YCbCr color space before and after smoke removal, subtracts and carries out binaryzation, further removes some static background interference area thereby obtains the suspected smoke area. Finally it extracts the Y channel of YCbCr color space on the suspected smoke area before and after smoke removal, transforms to two-dimensional discrete wavelet, and makes difference according to the wavelet energy, thereby obtains the final smoke area. Experimental results show that the proposed algorithm can effectively eliminate the interference of the color suspected smoke region in the image, and detect the smoke area accurately.