Image Caption And R-tree Index Optimization Method Of Civil Engineering Supervision

The rapid development of civil engineering has promoted the development of civil engineering supervision(CES)industry,and CES plays an important role as an important mechanism to ensure that civil engineering complies with relevant regulations.Image data is one of the main data generated during the CES process.In actual CES,due to the complexity of the scene,etc.,the use efficiency is not high.This paper focuses on these issues,the main research work is as follows.The first is to solve the problem that the CES image is not effectively labeled,and the parameter number is too much,which leads to the inefficiency in the production environment.Based on the existing work,this paper proposes the attention map model method based on Kmax mapping,relying on the attention mechanism.The recognition advantage of the key content of the image is achieved by mapping the attention-related layer in the image processing model based on the Kmax function,and then transfer the mapping result to improve the overall transfer effect of the model,thereby reducing the effect of the model as much as possible;then the related algorithm of the model transfer method based on Kmax mapping applied to the image caption model is given.Under the premise of ensuring the basic effect of the image caption model,the volume of the model is reduced,and finally the relevant caption model is applied to the civil engineering.At last,the analysis of the caption effect was carried out.Secondly,aiming at the problem of image indexing and retrieval efficiency of CES,combined with relevant data,the basic division of CES scenes and events is given,and based on this,the semantic engineering image data of the CES is divided into semantics;On the basis of semantic division,the semantic segmentation is processed based on the word vector,and the R-tree node splitting algorithm based on spectral clustering optimization is proposed.The algorithm is used to improve the indexing efficiency of these data.After the R-tree and the semantically divided annotation data,this paper proposes R-tree node retrieval optimization method based on word vector.Finally,the image classification model transfer algorithm proposed in this paper has a maximum loss of 1.6% in the CIFAR dataset compared with the reference model.The image caption model transfer method proposed in this paper has a maximum increase of 0.7% on the CIDEr benchmark in the MS COCO dataset.For the annotated CES image dataset,the R-tree based on spectral clustering optimization node splitting algorithm is reduced by about 11% in index time.On this basis,the word vector optimization based R-tree node retrieval algorithm is optimized to a maximum of 13% in the recall rate compared to the original R-tree.The above experimental results show that the proposed algorithm can improve the processing efficiency of CES image content to a certain extent,and it has certain feasibility.