Abstract: Weld infrared on-line monitoring system was established for gas metal arc welding. A real-time detection algorithm of infrared weld saliency based on frequency domain was proposed. The frequency tuned（FT） algorithm was improved from three aspects of filter processing,color model and background suppression,so as to obtain better infrared weld saliency map. The improved FT algorithm was compared and analyzed with some typical saliency detection algorithms such as ITTI,GBVS,SR,FT,AC and HC. The results showed that the average running time of the improved FT algorithm was 0. 075 s and no preprocessing was required,which effectively guaranteed the real-time performance of the detection process. Each algorithm was used to process the infrared image of the weld. The results showed that the improved FT algorithm could effectively suppress interfering pixels,such as background,noise and arc in the saliency map of the infrared weld,and also eliminate the interference of the heat affected zone. At the same time,the outline of the target weld in the saliency map was clear and the saliency was enhanced,which was beneficial to the feature extraction of the weld.