ZEMCH 2019 International Conference Proceedings April.2020 | Page 371
3. Results
3.1. Result of S‐CNN Model Development and Output of Test
As a result of validation S‐CNN model based on the dataset collected in the method section, the
model with the accuracy of 0.78 after 50epochs was achieved as shown in Table 3. Using the
developed model, interior wall of a speficif building was photographed using a smartphone, and the
WBS and progress status of the digital image were extracted as the output (Figure 3.).
Table 3. Result of S‐CNN Model Validation
Input
Number
of Data
WBS Level 1
Total: 13,442
Training: 11,855, Validation: 1,587
(Random Distribution)
Data
Structural Wall, Architectural Wall
Categories
Result
(WBS2)
WBS Level 2
Total: 13,442
Training: 9,404, Validation: 4,038
(Random Distribution)
Concrete Wall, Masonry Wall, Tile Wall,
Drywall
Epoch 1/50 9404/9404
[==============================] - 64s 7ms/step -
loss: 6.1143 - acc: 0.5641
Epoch 2/50 9404/9404
[==============================] - 64s 7ms/step -
loss: 7.1246 - acc: 0.5556
Epoch 3/50 9404/9404
[==============================] - 63s 7ms/step -
loss: 7.1246 - acc: 0.5556
…
Epoch 48/50 9404/9404
[==============================] - 72s 8ms/step -
loss: 0.0259 - acc: 0.99886
Epoch 49/50 9404/9404
[==============================] - 69s 7ms/step -
loss: 0.0939 - acc: 0.9786
Epoch 50/50 9404/9404
[==============================] - 68s 7ms/step -
loss: 0.015 - acc: 0.9973 3135/3135
[==============================] - 7s 2ms/step
loss= 1.09406523 accuracy= 0.777070091
(Bar chart: Loss, Dots: Accuracy )
(a) (b)
Figure 3. The result of tests and extracted information from digitalized pictures: (a) Digital
photographs and outputs of S‐CNN model in html. format; (b) Positioning information from picture’s
metadata.
Computer Vision and 4D BIM Applied Construction Project Progress Monitoring
360