Deep Learning-Based Performance Enhancement for Autonomous Excavator Systems

Automation and digitization in the construction industry are rapidly advancing, with a significant focus on the automation of excavators to enhance productivity and safety.This research presents a novel approach to improve the efficiency and stability of excavator operations by developing a vision-based autonomous excavator system.Given the complexity of real-time construction sites, accurate predictive models are crucial.While ANN (Artificial Neural Network) models are widely used, their predictive accuracy can be influenced by various factors such as input variables, network architecture, and learning methodologies.This study introduces an efficient deep learning-based ANN training model for intelligent automatic excavation systems, with a particular focus on optimizing input variable processing and model architecture.The research involved extensive experimental tests using diverse video data from construction sites, revealing that the choice of normalization methods for input variables significantly impacts the model's predictive performance.The findings highlight that ensemble models, which integrate multiple input variables, significantly outperform individual models.Specifically, non-linear normalization methods (Norm_3) showed superior predictive accuracy and reliability compared to linear methods (Norm_1 and Norm_2), with Norm_3 achieving a 91.0%accuracy and the combined model of Norm_2 and Norm_3 reaching an even higher accuracy of 92.1%.These results underscore the importance of selecting appropriate normalization techniques and the potential of multi-model collaboration in enhancing predictive accuracy.The findings not only advance the understanding of model optimization in autonomous excavation but also provide valuable insights that can enhance the performance of various AI(Artificial Intelligence) models in real-world construction automation applications.This study significantly contributes to the field of construction automation by offering a robust framework for improving the predictive performance of intelligent excavator systems, ultimately promoting safer and more efficient construction practices.