Prediction of concrete slump by RGB-D image feature fusion

Slump is an important index for concrete fluidity, which has a direct guiding effect on construction. In recent years, using RGB images for evaluating slump has been confirmed by scholars. Based on previous studies, this paper investigates the superiority of RGB-D image data over RGB image data in predicting slump of concrete and proposes three RGB-D fusion models: The early-stage-fusion model performs feature fusion in the data input stage, while the fully-connected-layer-fusion model performs feature fusion in the classification layer and the middle-stage-fusion model performs feature fusion after each residual block. In the classification of slump 120 mm, 150 mm and 200 mm, the Precision, Recall and F1-score are used to evaluate the model's ability to classify a single class, and the Accuracy, Macro-F1, Kappa and MCC are used to evaluate the model's performance. The experimental results showed that compared with the model using only RGB images, the fusion model achieve better performance, indicating that RGB-D image data can better evaluate concrete slump.