Industrial instrument detection method based on fusion of domain adaptation and unsupervised technology

The invention relates to an industrial instrument detection method based on domain adaptation and unsupervised technology fusion. The method comprises the following steps: step 1, constructing a source domain data set composed of generated images and a target domain data set composed of real shot images; 2, building a target detection network; 3, obtaining the confidence coefficient of bounding box regression and the overall output confidence coefficient; 4, training the target detection network in the step 2; step 5, simultaneously utilizing the source domain data and the target domain data to carry out domain adaptation training on the target detection network obtained in the step 4, and selecting the target domain sample and the source domain sample with the highest confidence coefficient to carry out splicing to obtain a spliced image and a pseudo label thereof; step 6, calculating the consistency loss between the spliced prediction result in the step 5 and the pseudo label; 7, calculating the total loss