Abstract: Reliable online penetration detection is an important issue in the intelligent manufacturing of laser welding, but the penetration signal belongs to the mesoscopic scale optical signal, and traditional detection methods are constrained by macroscopic sampling methods, which cannot effectively capture mesoscopic penetration information. Therefore, it is of great significance to carry out a study of the deeper mesoscopic properties within the laser keyhole to achieve reliable online penetration detection. In this paper, a mesoscopic penetration signal in the infrared spectrum was identified using a specialized optical imaging system. The study revealed that this signal possesses distinct physical characteristics, including strong concealment, high directionality, large fluctuation, and significant trend. Especially, it has been fully proven that the trend characteristics of the mesoscopic penetration signal under complex fluctuating surface. Through the abstract extraction and comparison analysis, it was found that the features are not only repeatable and subject to the nature of welding thermal reactions but also in good agreement with the laser welding penetration state. This study can provide crucial theoretical support for the development of reliable online detection of laser penetration at the mesoscopic scale.