The invention discloses a high-precision
laser welding spot recognition method based on an attention guidance context fusion network. The method mainly solves the problems that according to an existing
laser welding spot recognition technology,
small target features are prone to being lost, multi-scale detection is unbalanced, and
background noise interference exists. The implementation scheme is as follows: 1) acquiring a
data set and an identification tag; 2) constructing a
laser welding spot identification model; 3) constructing a
loss function; 4) training a laser welding spot identification model; and 5) obtaining a laser welding spot identification result. According to the laser welding spot recognition model constructed by the invention, the problem of multi-scale defect coexistence is solved by adopting differential cavity
convolution to capture local details and a semi-
global structure through the multi-scale context fusion module, the overlapping space reduction and local
convolution enhancement technology is adopted through the local enhancement space attention module, and the laser welding spot recognition accuracy is improved. The problem of insufficient retention of key edge information in deep features is solved, the attention guide dynamic sampling module replaces traditional
pooling operation to reduce down-sampling
information loss, the problem of low
retention rate of
small target features is solved, a channel and a space attention mechanism are fused in shallow features through the double-attention feature modulator module, and the target feature
retention rate is improved. The problem that the
small target detection rate is low under
background noise interference is solved.