Fluorescent magnetic particle inspection is widely applied in industry production, however; traditional ways are manual and depend on individuals' experience, which can also cause damage to workers' eyes. An automatic fluorescent magnetic particle inspection device is presented in this paper. A series of image processing methods have been investigated, which are put into a combination for detecting flaws in images automatically. And we propose a wavelet-based Canny operator to extract outlines of flaws in images, which is more effective than common Canny operator, and could constrain noises at a reasonable level. Results suggest that this novel automatic inspection device offers a low labor intensity but more effective way for flaw detection work, and image processing methods proposed in this paper are feasible.