Abstract: Visual detection of the state of the molten pool during the welding process is an important means to realize the online monitoring of weld quality. Aiming at the problems of molten pool unstable state and forming defects that are likely to occur during the climbing tungsten helium arc welding process of medium and thick aluminum alloys, this paper proposes a Tungsten Inert Gas Welding(TIG) welding status monitoring method based on the image characteristics of the molten pool. Based on the constructed passive vision sensor system, the acquisition of clear images of the molten pool under the interference of strong arc light is realized. A helium arc welding based on Otsu’s threshold segmentation and visual saliency features(VSF) is proposed. The image processing algorithm of the molten pool is used to extract the morphological features of the molten pool, and the relationship between the extracted visual features and the stability of the aluminum alloy climbing-TIG welding process is analyzed. Finally, a support vector machine (SVM) model is established to identify the welding state. The experimental results show that, compared with the geometric characteristics of the molten pool contour, the morphological characteristics of the molten metal at the end of the molten pool can more effectively reflect the unstable state of the molten pool during the aluminum alloy climbing-TIG welding process. The established welding state classification model has a maximum accuracy of 95.94% under the condition of a single feature input. The proposed real-time detection method provides a basis for online intelligent diagnosis and process optimization of TIG weld forming defects of large aluminum alloy components.