Deep Learning Algorithms in Predictive Models for Image Optimization Methods for Civil Engineering Supervision

The introduction of video surveillance in engineering supervision can enable remote monitoring of construction sites. By analyzing civil engineering supervision images through deep learning algorithms, a large amount of real-time image information can be obtained at the construction site, which can be used as a record of the work of the site supervisor and provide a basis in case of construction quality disputes. The objective of this paper is to study the prediction model for the optimization method of civil engineering supervision images based on deep learning algorithms. Methods for convolutional neural networks and image optimization are presented. The MP-SSD algorithm is introduced to deal with the defects of the SSD model, including the differentiation of the degree of contribution of the feature map by weighted feature fusion and the improvement of the rich semantic information. Deeply separable convolutions are introduced to reduce the number of model parameters and to improve inference speed. MP-SSD was tested and verified to be more accurate than the SSD model.