Automated Visual Quantification of CPT-like Penetration Responses in Transparent Soil Using Efficient Semantic Segmentation

Abstract High-resolution transparent-soil penetration sequences provide direct visual records of probe motion and local bubble/cavity development, but their quantitative interpretation remains difficult because the probe and scale occupy slender image regions, bubble/cavity boundaries are weak and irregular, and reflections from the model box and pore fluid frequently interfere with target regions. This study presents an automated engineering image-analysis workflow that combines four-class semantic segmentation, scale calibration, connected-component filtering, and geometric fitting for repeatable extraction of penetration-related visual indicators. A previously developed lightweight MambaDUPer framework was reconfigured to distinguish the background, probe, scale, and bubble/cavity regions in a dataset of 3,199 transparent-soil images. The implementation used a class-imbalance-aware loss for the sparse bubble/cavity class. The predicted masks were converted into penetration depth, laterally projected bubble/cavity area, and probe inclination. On the validation set, MambaDUPer achieved a mean intersection over union of 92.5%, an average accuracy of 99.6%, a mean recall of 96.4%, and an inference speed of 240.4 frames per second. Its inference speed was 5.40, 2.76, and 1.51 times that of SegFormer, the MobileNetV2-based model, and Fast-SCNN, respectively. The automatically extracted penetration depths agreed well with manual measurements, with a mean absolute error of 3.33 mm, a root mean square error of 3.52 mm, a mean absolute percentage error of 4.13%, and a coefficient of determination of 0.995. The extracted area–depth and inclination–depth curves captured staged changes in the visible bubble/cavity regions and probe attitude. The main contribution is a task-specific segmentation-to-measurement workflow for repeatable batch analysis of transparent-soil CPT-like image sequences. The extracted quantities are interpreted as image-response indicators under consistent imaging and calibration conditions rather than direct measurements of cone resistance, pore pressure, three-dimensional cavity volume, or complete soil failure zones.

Authors

Institutions

Publication Details

Journal
Engineering Research Express
Published
2026-11-17
DOI
https://doi.org/10.1088/2631-8695/aec35a
Primary Topic
Geotechnical Engineering and Soil Mechanics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Automated Visual Quantification of CPT-like Penetration Responses in Transparent Soil Using Efficient Semantic Segmentation

Yuntao Zhu, Mingfei Zhang, Yuntong Liu
2 links
Engineering Research Express
Geotechnical Engineering and Soil Mechanics
article

Automated Visual Quantification of CPT-like Penetration Responses in Transparent Soil Using Efficient Semantic Segmentation

Yuntao Zhu, Mingfei Zhang, Yuntong Liu
article en

Abstract

Abstract High-resolution transparent-soil penetration sequences provide direct visual records of probe motion and local bubble/cavity development, but their quantitative interpretation remains difficult because the probe and scale occupy slender image regions, bubble/cavity boundaries are weak and irregular, and reflections from the model box and pore fluid frequently interfere with target regions. This study presents an automated engineering image-analysis workflow that combines four-class semantic segmentation, scale calibration, connected-component filtering, and geometric fitting for repeatable extraction of penetration-related visual indicators. A previously developed lightweight MambaDUPer framework was reconfigured to distinguish the background, probe, scale, and bubble/cavity regions in a dataset of 3,199 transparent-soil images. The implementation used a class-imbalance-aware loss for the sparse bubble/cavity class. The predicted masks were converted into penetration depth, laterally projected bubble/cavity area, and probe inclination. On the validation set, MambaDUPer achieved a mean intersection over union of 92.5%, an average accuracy of 99.6%, a mean recall of 96.4%, and an inference speed of 240.4 frames per second. Its inference speed was 5.40, 2.76, and 1.51 times that of SegFormer, the MobileNetV2-based model, and Fast-SCNN, respectively. The automatically extracted penetration depths agreed well with manual measurements, with a mean absolute error of 3.33 mm, a root mean square error of 3.52 mm, a mean absolute percentage error of 4.13%, and a coefficient of determination of 0.995. The extracted area–depth and inclination–depth curves captured staged changes in the visible bubble/cavity regions and probe attitude. The main contribution is a task-specific segmentation-to-measurement workflow for repeatable batch analysis of transparent-soil CPT-like image sequences. The extracted quantities are interpreted as image-response indicators under consistent imaging and calibration conditions rather than direct measurements of cone resistance, pore pressure, three-dimensional cavity volume, or complete soil failure zones.

Engineering Research Express
Zhengzhou University of Aeronautics (CN)
National Natural Science Foundation of China
Geotechnical Engineering and Soil Mechanics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon