초음파를 이용한 CNN 기반 다중 굴곡 관 환경 수위 측정

초록

A CNN-based classification model for water level measurement in a multi-bent pipeline environment using ultrasonic signals is constructed. Conventional water level measurement methods are often designed for straight structures or simple environments, and their accuracy may degrade in pipeline environments with bends or obstacles due to signal scattering and multiple reflections. To address this problem, a method was applied in which received ultrasonic signals were converted into image representations and water levels were classified using a CNN. Experiments were conducted under two conditions: an environment without obstacles and an environment with obstacles. The performance was compared with AlexNet, GoogLeNet, ResNet 18, ResNet 50, DenseNet 201, and ViT models. In addition, a one-stage serial CNN structure was constructed by adjusting the filter size and stride combinations of the convolutional layer, and the performance of various combinations was analyzed. Experimental results show that the constructed model achieved high classification accuracy with shorter training time compared with existing CNN models. These results confirm that CNN-based methods can be effectively utilized for water level measurement in a multi-bent pipeline environment.

키워드