다중이용시설

다중이용시설

結果 (繁體中文) 1:

多用途設施

結果 (繁體中文) 2:

多功能設施

結果 (繁體中文) 3:

In tunnel environment, the inuences of background noise, such as background textures, lamp-shading, and manual marks on the surfaces of the tunnel linings, tend to increase the diversity of the data and pose major challenges for the accurate identication of damages.Fig.12 expressed the images which had been influenced by the background noise in the current study.Figs.12b, 12c, and 12d illustrated that the proposed method had displayed excellent prediction results for the images with variable background textures and manual marks.However, it was observed that the basic SegNet method could not completely identify the profiles of the cracks and spalling (Figs. 12b and 12c), and it had also failed to recognize the cracks detailed in Fig.12d.Although the two-stream method was competent for accurately identifying the profiles of the spalling (Fig. 12c), it had unfortunately obtained unsatisfactory recognitions of the crack damages, as depicted in Figs.12b and 12d.Furthermore, as shown in Fig.12a, the prediction results of the proposed method showed excellent agreement with the ground truths.Meanwhile, both the two-stream method and the basic SegNet method displayed difculties in extracting the detailed features of the damages.Therefore, both the two-stream method and the basic SegNet method underperformed when compared with the results achieved by the proposed FL-SegNet in regard to the images with background interference.