In order to measure the temperature of radiation image target，the measured target must be recognized accurately from radiation images in the non-contact soft measurements of temperature field based on CCD image sensor. It is difficult to recognize the image target because of the existence of various interferences in radiation images captured in industrial locale. A classification recognition method is proposed. By multi-spectrum segmenting，various high-temperature noises in the radiation color image are reduced or even eliminated. And then with the improved Ostu segmentation algorithm，the interference of smog is eliminated. Finally，the morphology method in mathematics is applied to process the segmented image and remove the dissociations and narrow holes to smooth the image’s edge. The experimental results show that the method can recognize high-temperature melt target from high-temperature radiation image with various interferences accurately and it has an excellent practicability.