Image Semantic Generation In Intelligent Manufacturing

With the rapid development of intelligent manufacturing technology,the workshop equipment in the factory is gradually intelligent,and the quality of video monitoring data in the workshop is getting higher.How to use a large number of image and video data combined with artificial intelligence technology to realize intelligent management of factories is an urgent task.Combining image semantic analysis and intelligent manufacturing can realize intelligent monitoring of production process.This paper mainly focuses on how to apply the image semantic analysis to the intelligent manufacturing and place emphasis on the visual attention mechanism and the analysis of tagging sentence attributes.The main work is summarized as follows:(1)Aiming at the problem that the background of the production workshop is chaotic.By using the ability to extract the foreground feature of image by using the target detection model,a Visual-Semantic Attention(VSA)model based on top-down Attention is proposed.VS A model introduces the visual attention based on the basic framework.Firstly,the target detection network based on YOLOv3 is used to extract the visual features in the image and encode them.Secondly,the language model based on the two-layer long and short term memory network is used to generate the final production image annotation.At the same time,in the training process,in order to improve the accuracy of the model without reducing the efficiency of the model,a transfer learning algorithm training network was designed.In order to better verify the performance of the model in the actual production process,an industrial dataset-XRindustry based on the actual industrial production was built.The experimental results on XRindustry verify that the visual attention module can effectively improve the personality of the model and enhance the application of the model in the intelligent manufacturing.(2)Aiming at the problem of low relevance between generated words by traditional language models and visual features of images,an automatic image annotation model based on attribute analysis(Semantic Attention,SEA)was proposed by taking advantage of the attention mechanism in natural language processing.In order to make the generated annotation contain more image information,SEA model firstly uses pointer network and visual weight to analyze the visual and semantic characteristics of each word,and then generates a complete annotated sentence.The experiment results of the model on MSCOCO and XRindustry verify that the design of attribute analysis module can effectively improve the relevance between production description and image features,and verify that the addition of visual model-based attention mechanism in the language model can effectively improve the coherence and word order accuracy of generated sentences.