With the rapid development of society, the demand for steel production is increasing, and the demand of the bar’s quality is getting higher and higher. However, what is constraining the production efficiency of the Bar industry is that the development of NDT of surface crack on the bar is unpleasant. This paper presents a new NDT method which is machine vision-based and fully automatic, and designs an application system which is more viable and practical. The bar non-destructive crack detection system use the Industrial Control Computer as Master Parts. Combine with the CCD image acquisition and the PLC control of the pipeline as a whole. This system utilizes Testing Technology of Fluorescent Magnetic particle, realize machine recognition through the Digital Image Processing Technology. What’s more, this system use the set method of adaptive threshold based on probability of detection and false alarm probability in machine recognition. It has a significant and Practical significance for reducing the computation of the system, improving the processing speed of this system, and meeting the needs of the entire system in real-time requirements.