These four datasets collectively serve as benchmarks for surface crack detection across road pavements and stone materials, distinguished by their varying imaging technologies and preparation methods. The CrackTree260 dataset focuses on deep learning readiness, expanding 260 visible-light, area-array pavement images into a massive training set of 35,100 samples through rigorous augmentation. In contrast, CRKWH100 and CrackLS315 prioritize high-resolution precision using line-array camera...

Over-excavation by Earth Pressure Balance (EPB) shields results in ground loss, which can lead to surface subsidence and pose a serious threat to both infrastructure and transportation systems. Current methods for monitoring muck volume—typically relying on weighing or visual inspection of the muck box—are often time-consuming, prone to error, and lack reliability. To address this, this paper proposes a muck volume measurement algorithm combined with a 3D point cloud segmentation model, ...

A large language model capability evaluation set containing 200 Q&A pairs in the field of hydraulic structures. This is a large language model capability evaluation set containing 200 questions in the field of hydraulic structures. The evaluation set comprehensively covers major sub-disciplines of hydraulic engineering including hydraulic tunnels, dams, hydropower stations, sluice gates, and flood discharge facilities in terms of content, and evenly includes four question types: multiple...

This dataset has (i) 3D geometry files of mineral particles from Florida and Virginia and (ii) analyzed 3D particle geometries. A total of 382 particle geometry files are included in this dataset. The Florida particles in this dataset are limestones, while the Virginia particles are freshly crushed granites from a quarry in Richmond, Virginia. The geometry files are obtained by 3D structured light scanning and presented in wavefront .obj format. These files can be viewed using free 3D ge...

This folder contains 48 images of grains and the corresponding segmentation masks that form the majority of the images that were used to train the U-Net model that the 'Segmenteverygrain' Python package is based on. The images have filenames that terminate in '_image.png'; the mask filenames terminate in '_mask.png'. The mask rasters only contain three values: 0 for background, 1 for the grain itself, and 2 for the grain boundary.

The dataset consists of two primary components: the data itself and the implementation code for constructing the multi-fidelity model. The data component comprises two types of sources: low-fidelity data derived from chart-based knowledge, and high-fidelity engineering data collected from actual shield tunneling projects. Both types of data include shield tunneling parameters along with the corresponding geological parameters.

This dataset consists of two types of image data for sand particle analysis, covering a variety of typical sand types. It is suitable for tasks such as image super-resolution, object detection, and instance segmentation. Sand Particle Super-Resolution Dataset contains 800 high-resolution images (4480×4480 pixels) captured using a Canon 5D Mark IV camera equipped with an EF 100 mm f/2.8 L IS USM macro lens. The dataset includes 8 types of sand particles: Fujian sand, Hebei Sand No.1, Heb...

This dataset is designed to support the automatic recognition and segmentation of nonoverlapping and overlapping particles in granular images. All images were captured from a top-down view using a Canon 5D Mark IV camera equipped with an EF24-105mm lens, with a resolution of 6720×4480 pixels. The dataset contains a total of 420 images. Each image has been annotated at the instance level using the LabelMe tool, with particles categorized into two types of labels: “nonoverlapping particles...

The datasets (84 flume, 118 field photos) cover a wide range of fluvial environments and include a variety of field site and flume experiment images. The datasets were grouped into three subsets according to sediment and channel conditions: (1) flume channel (84 photos; photo size ~ 0.2 m × 0.2m); (2) forested mountain rivers (70 photos; photo size ~ 1m × 1m); and (3) sparsely vegetated large rivers (6 photos; photo size ~ 20m × 20m). To train the machine learning model to better disting...

Database and code to the paper: A multiscale CNN-based intrinsic permeability prediction in deformable porous media. Including: (1) a dataset of CT images from Bentheim sandstone at different volumetric strain levels; (2) pore-scale permeability simulation data of single-phase flow using the lattice Boltzmann method (LBM); (3) the CNN model with the processed CT images as inputs and permeability tensors as outputs; and (4) data augmentation and alternative CNN architectures.