AWFMFNet image classification network model for concrete collapse

Concrete collapse is a quantitative indicator of the degree of fluidity of the mix, which is used to determine whether the construction can be carried out properly. During the whole cycle of concrete, it needs to go through three processes: preparation, transportation and pouring. In this paper, we...

Full description

Saved in:
Bibliographic Details
Main Authors: Zheng, Xin, Yan, Hairong, Gao, Xiao, Tu, Yubo
Format: Conference Proceeding
Language:English
Subjects:
Adaptation models
Adaptive feature fusion
Computer Vision
Concrete collapse degree
Deep learning
Feature extraction
Fuses
Image classification
Prediction algorithms
Predictive models
Transportation
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Concrete collapse is a quantitative indicator of the degree of fluidity of the mix, which is used to determine whether the construction can be carried out properly. During the whole cycle of concrete, it needs to go through three processes: preparation, transportation and pouring. In this paper, we focus on the transportation process of concrete and use computer vision algorithms with deep learning convolutional neural networks to determine the collapse value of freshly mixed concrete inside a mixer truck. To address the problem of how to determine concrete collapse easily, quickly and accurately, and to improve the efficiency of concrete collapse determination, this paper proposes a new structural model of AWFMFNet image classification network for concrete collapse. The network model designed in this paper enhances feature extraction and fusion by adaptive weighting algorithm and pixel-level feature enhancement algorithm, which enables the network model to further extract key features and fuse deep and shallow image features to better guide the image classifier to make final classification prediction. This paper conducts experiments using concrete collapse datasets collected from real application scenarios and compares them with several other classical image classification network models. The experimental results show that the network model can effectively extract the multilayer features of concrete images and has better collapse classification results. The network model improves the accuracy by 4.62% compared with ResNet50, 3.27% compared with DarkNet53, 1.74% compared with GoogLeNet, and 7.21% compared with VGG16.
ISSN:2693-2776
DOI:10.1109/IMCEC55388.2022.10020095