Page 25 - Kỷ yếu hội thảo khoa học lần thứ 12 - Công nghệ thông tin và Ứng dụng trong các lĩnh vực (CITA 2023)
P. 25
Van Vy and Hyungchul Yoon 9
where denote for predicted coordinate and true coordinate respectively, and n
denotes the number of samples in the experiment.
MAE measures the average absolute difference between the predicted values and
the true values, while RMSE measures the standard deviation of the differences
between the predicted values and the true values. It is obvious that the smaller the
MAE and RMSE, the better accuracy.
As shown in Table 2, the AECWT-3DR-Net model produced progressive results
on the dataset. The MAE on the x, y, and z-axis is less than 7.6 mm. The RMSE on
the x, y, and z-axis is less than 9.6 mm. The visualization of the test data is also
illustrated in Figure 8.
We also compared our proposed method with two other methods [6, 11] in Table 3.
The first method used single-input images to feed the CNN network. It is clear from
the results that using only a single input leads to significant errors due to a lack of
information. The second method, ToA, is a traditional method that is commonly used
in physical theory to localize the crack coordinates. This method gives better results
than the previously mentioned method, but errors are still largely attributed to the
attenuation of the signals and the inhomogeneity of the material. Specifically, when
utilizing information from all eight sensors (AECWT-3DR-Net), the errors were
smaller compared to the other methods.
Fig. 8. The testing results of the concrete cube were visualized in four planes
Table 2. The evaluation of the proposed method
AECWT-3DR-Net
Axis x y z
MAE (mm) 7.6 6.4 6.8
RMSE (mm) 9.6 7.9 9.3
ISBN: 978-604-80-8083-9 CITA 2023