Efficient and Interpretable Pavement Crack Classification via SylphNet: Systematic Hyperparameter Analysis on a Primary Road Image Dataset

ABSTRACT Automated pavement crack classification is essential for intelligent transportation infrastructure, yet existing deep learning methods often suffer from high computational cost, limited interpretability, and insufficiently justified hyperparameter selection. This study proposes SylphNet, a lightweight convolutional neural network (CNN) for multi‐class pavement crack classification, supported by an ablation‐driven experimental framework. The primary dataset originally contained 12,000 real‐world pavement images collected in Pabna, Bangladesh, representing edge cracks, alligator cracks, and crack‐free surfaces. After exact‐duplicate auditing, five duplicate images were removed, resulting in 11,995 images for experimentation. Ablation studies investigated input resolution, data augmentation, learning rate, and dataset split. An initial split comparison showed that an 80:20 train–test split improved accuracy by 0.41% over a 75:25 split, while the final protocol employed separate training, validation, and locked‐test sets. SylphNet integrates depthwise separable convolutions and residual bottlenecks, achieving 97.50% accuracy with only 1.036M parameters and approximately 2.1 ms inference latency per image. Comparisons with DenseNet169, MobileNetV2, EfficientNet‐B0/B1, AlexNet and VGG16/19 demonstrate its competitive performance and computational efficiency. The study further employs repeated‐seed validation, computational profiling, and explainable AI techniques, including Grad‐CAM, Grad‐CAM++, saliency maps and occlusion sensitivity, to provide complementary evidence of crack‐relevant attention and model interpretability.

Authors

Institutions

Publication Details

Journal
Artificial Intelligence for Engineering
Published
2026-11-16
DOI
https://doi.org/10.1049/aie2.70031
Primary Topic
Infrastructure Maintenance and Monitoring
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Efficient and Interpretable Pavement Crack Classification via SylphNet: Systematic Hyperparameter Analysis on a Primary Road Image Dataset

Tamanna Yasmin, Shahriar Siddique Arjon, Nakib Aman, Md.Nayem Hossain et al.
Artificial Intelligence for Engineering
Infrastructure Maintenance and Monitoring
article

Efficient and Interpretable Pavement Crack Classification via SylphNet: Systematic Hyperparameter Analysis on a Primary Road Image Dataset

Tamanna Yasmin, Shahriar Siddique Arjon, Nakib Aman, Md.Nayem Hossain, Nymur Rahaman Antor
article en

Abstract

ABSTRACT Automated pavement crack classification is essential for intelligent transportation infrastructure, yet existing deep learning methods often suffer from high computational cost, limited interpretability, and insufficiently justified hyperparameter selection. This study proposes SylphNet, a lightweight convolutional neural network (CNN) for multi‐class pavement crack classification, supported by an ablation‐driven experimental framework. The primary dataset originally contained 12,000 real‐world pavement images collected in Pabna, Bangladesh, representing edge cracks, alligator cracks, and crack‐free surfaces. After exact‐duplicate auditing, five duplicate images were removed, resulting in 11,995 images for experimentation. Ablation studies investigated input resolution, data augmentation, learning rate, and dataset split. An initial split comparison showed that an 80:20 train–test split improved accuracy by 0.41% over a 75:25 split, while the final protocol employed separate training, validation, and locked‐test sets. SylphNet integrates depthwise separable convolutions and residual bottlenecks, achieving 97.50% accuracy with only 1.036M parameters and approximately 2.1 ms inference latency per image. Comparisons with DenseNet169, MobileNetV2, EfficientNet‐B0/B1, AlexNet and VGG16/19 demonstrate its competitive performance and computational efficiency. The study further employs repeated‐seed validation, computational profiling, and explainable AI techniques, including Grad‐CAM, Grad‐CAM++, saliency maps and occlusion sensitivity, to provide complementary evidence of crack‐relevant attention and model interpretability.

Artificial Intelligence for Engineering
Pabna University of Science and Technology (BD)
Infrastructure Maintenance and Monitoring
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