Weld joint quality detection method based on Transform neural network

The invention discloses a weld joint quality detection method based on a Transform neural network, which comprises the following steps of: S1, additionally arranging a high-speed capacity molten pool monitoring camera at the end, close to a welding gun, of a welding manipulator, acquiring welding shooting image data in real time, and monitoring the states of an electric arc, a molten pool, a groove and the like; s2, basic data enhancement is carried out on the acquired data, a multi-image overlapping enhancement method is used, and the detection efficiency of small targets such as welding seams is improved; s3, a visual Transform neural network welding seam recognition model based on self-attention is constructed; s4, visualizing the change of the training process of the model by adopting a multi-modal related attention method, and adjusting the structure of the model; and S5, detecting the quality of the welding seam by using the trained model. According to the method, the data has relatively high robustness