Dataset of soil particle images and particle size distribution for applications in granular materials Authors: Deep Arvinder, Gobinath Ravindran, Rutuja Gumathannavar Published: 2026 ABSTRACT AbstractThis dataset introduces soil particle images corresponding to sieve-based particle size distribution (PSD) obtained from black cotton soil sourced from the laboratory stock of the Geotechnical Laboratory, Symbiosis International (Deemed University), Pune, Maharashtra, India. A bulk sample of 1500 grams was processed by implementing the mechanical sieve analysis using ASTM/ISO standard sieves (2.36 mm to 0.075 mm) under mechanical vibration for 15 min in three stages of 500 grams each. Representative fractions from each sieve were used for multi-scale image acquisition. A total of 1350 smartphone images were captured with the use of an iPhone 15 Pro Max at three heights (15 cm, 30 cm, 45 cm), incorporating repeated disturbance cycles to ensure the variability in particle exposure. In addition to the smartphone images, 1350 microscopic images were acquired from sub-sampled sieve fractions using varying magnifications (4x, 10x, and 20x) under an inverted microscope Olympus CKX53. The inclusion of both macro-scale and micro-scale imaging aligned with standardized sieve data supports development and validation of predictive models in geotechnical engineering. Even though the dataset is generated from black cotton soil obtained from a single source, it provides a useful benchmark for developing and evaluating image-based characterization methods. Application to other soil types or granular materials requires additional validation.Keywords: Soil particle images, Sieve analysis, Image processing, Particle size distribution, Machine learning, Granular material, Black cotton soil, Deep learning dataset Specifications Table SubjectEngineering & Materials scienceSpecific subject areaSoil Particle Size CharacterizationType of dataRaw Smartphone Images (JPEG)Microscopic Images with embedded cellSens dimensional measurement overlays (TIF)Smartphone Image Parameters (CSV)Microscopic Image Parameters (CSV)Data collectionParticle Size Distribution was obtained by Mechanical Sieve analysis using ASTM/ISO specified sieves (2.36 mm-075 mm) for 1500 g of black cotton soil. Images were acquired using an iPhone 15 Pro Max under ambient indoor lighting near a window on a white background at heights of 15, 30 and 45 cm with repeated disturbance. Microscopic imaging was performed using Olympus CKX53 with DP28 Colour Camera on 5 sub samples of 1.18 mm-0.075 mm each at magnification of 4x, 10x and 20x accordingly excluding 2.36 mm soil sample due to limited field of view. Image parameters taken: Focal length, Aperture, Light Intensity, Shutter Speed, Height of Camera, Image Resolution, Mean Intensity and Pixel Count accordingly.Data source locationInstitution: Symbiosis International (Deemed University) (SIU), Pune, India Pin: 412115Data accessibilityRepository name: IEEE DataPortData identification number: 10.21227/azjw-q909Direct URL to data: https://ieee-dataport.org/documents/image-processing-dataset-soil-particle-size-identification-and-other-related-purposesAccess to the dataset is subjected to IEEE DataPort access policies and requires an IEEE DataPort Subscription to access.Related research articleNone Value of the Data •This dataset contains multi-scale soil particle images of black cotton soil at both macro and micro levels aligned with the standardized Particle Size Distribution (PSD) obtained from conventional soil particle size distribution techniques. This allows a direct comparison between conventional laboratory-based methods and image-based characterization techniques. This dataset contains multi-scale soil particle images of black cotton soil at both macro and micro levels aligned with the standardized Particle Size Distribution (PSD) obtained from conventional soil particle size distribution techniques. This allows a direct comparison between conventional laboratory-based methods and image-based characterization techniques. •The combination of images from both smartphone and microscopic images will allow researchers to analyze the effect of different image scales on the prediction results, which is particularly significant for predicting soil particle size distribution, which is particularly relevant considering the recent interest in using machine learning models for geotechnical engineering. The combination of images from both smartphone and microscopic images will allow researchers to analyze the effect of different image scales on the prediction results, which is particularly significant for predicting soil particle size distribution, which is particularly relevant considering the recent interest in using machine learning models for geotechnical engineering. •This dataset provides quantitative variability relevant to machine learning and computer vision applications. Particle size diversity is represented by sieve sizes spanning 0.075 mm - 2.36 mm for smartphone images and 0.075 mm – 1.18 mm for microscopic images. The dataset comprises 1350 smartphone images and 1350 microscopic images and is supported by multiple imaging scales, heights and illumination intensity. Additionally, independently collected microscopic sub-samples introduce intra-fraction variability in texture representation and particle morphology. These quantitative characteristics support the application of this dataset to machine learning, image analysis and computer vision tasks. This dataset provides quantitative variability relevant to machine learning and computer vision applications. Particle size diversity is represented by sieve sizes spanning 0.075 mm - 2.36 mm for smartphone images and 0.075 mm – 1.18 mm for microscopic images. The dataset comprises 1350 smartphone images and 1350 microscopic images and is supported by multiple imaging scales, heights and illumination intensity. Additionally, independently collected microscopic sub-samples introduce intra-fraction variability in texture representation and particle morphology. These quantitative characteristics support the application of this dataset to machine learning, image analysis and computer vision tasks. •The dataset can be used for the evaluation and development of image segmentation workflows following user-generated annotations or derived through unsupervised approaches such as thresholding, edge detection, morphological operations etc. The dataset can be used for the evaluation and development of image segmentation workflows following user-generated annotations or derived through unsupervised approaches such as thresholding, edge detection, morphological operations etc. •The data can be utilized for feature extraction using the raw smartphone (JPEG) and microscopic (TIF) images, PSD prediction using corresponding sieve fraction labels and sieve based PSD data, image segmentation framework development following segmentation approaches, acquisition condition analysis using the associated metadata (height of camera, focal length, aperture, shutter speed, light intensity and image resolution), classification, physics-informed modelling, supporting further research on the development of automatic soil particle characterization techniques. The data can be utilized for feature extraction using the raw smartphone (JPEG) and microscopic (TIF) images, PSD prediction using corresponding sieve fraction labels and sieve based PSD data, image segmentation framework development following segmentation approaches, acquisition condition analysis using the associated metadata (height of camera, focal length, aperture, shutter speed, light intensity and image resolution), classification, physics-informed modelling, supporting further research on the development of automatic soil particle characterization techniques. Background Particle Size Distribution (PSD) in geotechnical engineering is fundamental for evaluation of its engineering behavior and classification [1]. In addition to particle size, particle shape also influences its mechanical behavior and packing characteristics of granular materials [2]. PSD is conventionally done using standardized laboratory tests that include mechanical sieving and sedimentation procedure based on ASTM/ISO standards. These methods are time-intensive and labor-intensive specifically for fine-grained soils. This has influenced image-based characterization approaches including dynamic image analysis methods as it delivers improved efficiency and reduced human error compared to hydrometer-based techniques [3]. Recent studies in image processing and machine learning frameworks enabled alternative estimation of PSD using digital images [4]. Computer vision techniques adopted by standard mobile phone cameras have reported reasonable in estimating PSD compared to sieve analysis procedure [5]. Image-based techniques extract particle-level characteristics which can be correlated with sieve-based distributions [6], though their accuracy is dependent on image conditions which include illumination, contrast and noise control [7]. Automated and real-time imaging systems have also been adopted in industrial granular materials [8]. Deep learning approach further enhances image-based soil and aggregate characterisation [9,10]. Available datasets provide paired soil images and PSD data [11,12] promoting development of robust data-driven models for granular material application [13]. Data Description The dataset provides a detailed multi-scale representation of soil characteristics by combining macroscopic imaging, microscopic analysis and associated sieve-based particle size distribution data. It is organized structurally into two main categories, Smartphone Image Dataset and Microscopic Image Dataset along with corresponding sieve-based particle size distribution (PSD) data, allowing comprehensive assessment of soil structure, texture, and particle size variation across different fractions. Smartphone images are provided in JPEG (.jpeg) format and microscopic images are provided in TIF (.tif) format corresponding to the native output format of the camera systems. Image Parameter files are provided in CSV (.csv) format for machine-readable tabular data handling. PSD data is presented in tabular form within the manuscript. These formats primarily represent the default outputs of the acquisition systems and were retained to preserve the original data characteristics except for the microscopic images, which carry measurement overlays generated with the acquisition software cellSens. The smartphone image dataset contains 225 images across 6 sieve samples totaling 1350 images whereas microscope image dataset consists of 54 images per 5 sub-samples and same across 5 primary sieve samples totaling 1350 images. Smartphone image dataset From each sieve fraction, a representative 75 g of soil sample was placed on a plain white paper. First, a set of three images were captured at predefined camera heights of 15 cm, 30 cm and 45 cm. Following this, the soil arrangement was manually disturbed using a paper lifting and turning procedure. No predefined spreading procedure was adopted. Specifically, one corner or one side of the A4 sheet paper was lifted and rotated inward in the opposite direction to overturn the top particle layer. The procedure was then repeated using other corners or sides to ensure that previously underlying particles were brought to the surface while previously exposed particles were redistributed below. This disturbance procedure was aimed to generate variability in particle exposure and arrangement. The same procedure was adopted consistently across all samples and disturbance cycles. The process was repeated for 75 g sample from each sieve fraction separately throughout all 75 disturbance cycles. Upon completion of each disturbance cycle, another set of three images were acquired at the predefined heights, repeated throughout. The disturbance procedure was conducted manually; therefore, no quantitative force or time measurement was imposed. Table 1 describes the detailed parameters and dataset information acknowledged for the smartphone image dataset. Table 1Dataset information for smartphone-based image.Table 1: dummy alt textSmartphone-based ImagesSoil typeBlack CottonSoil Mass per Sieve Sample75 gNumber of Sieve Samples6Sieve Sample Sizes(2.36, 1.18, 0.6, 0.3, 0.15, 0.075) mmCamera Heights per Disturbance Cycle15 cm, 30 cm, 45 cmImages Captured per Disturbance Cycle3Disturbance Cycles per Sieve Sample75Images per Sieve Sample225 JPEG ImagesTotal Number of Smartphone images1350Image Name StructureSoilType_Month_Batch_ImageName Dataset information for smartphone-based image. The image structure is framed as per the type of soil, month of acquisition, batch number and a unique image identifier (e.g., BlackCotton_March_1_IMG_1227). Image specific metadata and parameters including the corresponding sieve size, focal length, aperture, shutter speed, height of camera, light intensity and image resolution are mapped to each image in the form of accompanying Smartphone Image Parameters CSV file using the image name structure as the unique identifier The filename format does not encode the sieve fraction, camera height, disturbance cycle or image sequence. The image-specific parameters are associated with each image through the accompanying Smartphone Image Parameters CSV file using image name as a unique identifier. Microscopic image dataset Microscopic images of soil samples were generated using an inverted microscope, Olympus CKX53 equipped with a DP28 color camera. The camera features a resolution of 8.9 megapixels, providing high-quality visual representation of soil particles. A total of 1350 images were captured. To ensure systematic data organization and traceability, microscopic images follow the same image name structure as the smartphone image dataset and are shown in Table 2. Key imaging parameters considered during the image acquisition including sample identifier name, sieve size, mean intensity, and fixed camera height, were carefully controlled and recorded in the accompanying Microscopic Image Parameters CSV file to maintain consistency and reproducibility throughout the dataset generation process. Table 2Dataset information for microscope-based image.Table 2: dummy alt textMicroscope-based ImagesSoil TypeBlack CottonNumber of Sieve Samples5Sieve Sample Size1.18 mm0.6 mm0.3 mm0.15 mm0.075 mmCorresponding magnification value4x4x10x20x20xSub-Sample per Sieve Sample5Images per Sub-Sample54Total Images per Sieve Sample270 TIFF ImagesTotal Number of Microscopic Images1350Image Name StructureSoilType_Month_Batch_ImageName Dataset information for microscope-based image. The microscopic image file name structure is framed following the same format as the smartphone image dataset. The image structure is framed as per the type of soil, month of acquisition, batch number and a unique image identifier (e.g., BlackCotton_March_1_IMG_1227). Image Number will vary in accordance with the image. The dataset is classified into two main image repositories: the Smartphone Image Dataset and the Microscope Image Dataset, each organised based on the particle size fractions that were determined by sieve analysis. The hierarchical organization of the dataset is summarized in Table 3. Fig. 1. Table 3Folder dataset structure shown in table below.Table 3: dummy alt textPrimary FolderDatasetSmartphone Image DatasetSubfolders2.36mm1.18mm0.6mm0.3mm0.15mm0.075mmPrimary FolderMicroscope Image DatasetSubfolders1.18mm0.6mm0.3mm0.15mm0.075mmSub-Sample Folders (A-E)1.18 A0.6 A0.3 A0.15 A0.075 A1.18 B0.6 B0.3 B0.15 B0.075 B1.18 C0.6 C0.3 C0.15 C0.075 C1.18 D0.6 D0.3 D0.15 D0.075 D1.18 E0.6 E0.3 E0.15 E0.075 EMetadata FileSmartphone Image Parameters (csv)Metadata FileMicroscopic Image Parameters (csv)Readme FileREADME (txt)License FileLicense CC BY 4.0 (txt)a: Example of Image Files in Image Dataset FoldersSmartphone Image DatasetBlackCotton_March_1_IMG_1623.jpegMicroscope Image DatasetBlackCotton_March_1_Image_1433.tif Folder dataset structure shown in table below. Schematic representation of the pipeline for the data collection and Dataset generation. Experimental Design, Materials and Methods Soil sampling A bulk soil sample of 1500 g of black cotton soil was collected from the laboratory stock of Geotechnical Laboratory at Symbiosis International (Deemed University), Pune, Maharashtra, India, in an already air-dried condition; therefore, no additional drying procedure was performed prior to testing. Also, moisture content was not measured prior to the sieve analysis of the soil sample. The sample was initially pre-sized through a 4.75 mm sieve to remove coarse fractions and to ensure a representative fine grained soil matrix. The processed soil sample was further subdivided into three specimens of 500 g each to facilitate efficient and accurate sieve analysis as processing of large sample masses may lead to particle congestion specifically in the finer sieves where soil accumulation may restrict aperture openings further hindering the passage of finer particles. In accordance with ASTM D6913/D6913M-17 and ISO 17892–4 standards, which recommend the use of appropriately sized representative specimens to prevent sieve overloading and ensure effective particle separation, the subdivision approach was adopted. Following sieve analysis procedure, 75 g of the retained soil corresponding to each sieve fraction was randomly drawn from each sieve and stored separately. Mechanical sieve analysis A total of 1500 g of black cotton soil was taken for sieve analysis procedure. The soil was sieved in three batches of 500 g to avoid overloading the finer sieves to ensure accurate particle separation. The procedure of mechanical sieving was performed using a set of ASTM/ISO standard sieves with mesh sizes of 2.36 mm, 1.18 mm, 0.6 mm, 0.3 mm, 0.15 mm, and 0.075 mm. The stacked sieves were mounted on a mechanical sieve shaker as shown in Fig. 2 and operated for 15 min to ensure effective size-based particle separation. Thereafter, the mass of soil retained on each sieve was accurately measured. These values were later used to calculate the mass retained, percentage retained, and percentage finer for each size fraction, enabling detailed characterization of the soil particle size distribution. The separate fractions were then utilized for further analysis such as macroscopic imaging using a mobile phone and microscopic inspection to assess particle morphology. Mechanical sieve shaker setup, used for particle size analysis of soil. To structurally characterize soil particle morphology across various size fractions, a dual-mode imaging approach combining smartphone-based macroscopic imaging and microscopic analysis was employed. This approach aided in multi-scale visualization, capturing both bulk aggregation behaviour and fine particle-level features as shown in Fig. 3. Inverted microscope system, Olympus CKX53, used for acquisition of microscopic images. PSD values included in the dataset were obtained by performing mechanical sieve analysis procedure. The PSD data represents experimentally derived measurements based on the mass of soil retained on each of the sieves. The data is represented in the form of PSD curve as shown in Fig. 4. From the retained soil mass, a standard table and PSD curve were generated by calculating specific values mentioned in Table 4. PSD Curve analysis of soil. Table 4Particle Size Distribution of 1500 g of soil sample.Table 4: dummy alt textSieve No.Particle Size (mm)Mass Retained (g)% Retained% Finer82.36865.7394.27161.1815310.2084.07300.624416.2767.80500.339526.3341.471000.1533822.5318.932000.07515810.538.40Pan1268.400.00Total1500 Particle Size Distribution of 1500 g of soil sample. No additional computational filtering or other post processing were applied beyond the standard sieve-based size classification. Smartphone image acquisition Smartphone images were acquired using an iPhone 15 Pro Max camera system equipped with a 48-megapixel primary camera representative images are shown in Fig. 5a. It was performed in daytime under indoor ambient lighting conditions near a window to utilize the natural daylight while avoiding direct sunlight exposure onto the soil sample. Direct solar illumination was avoided to minimise strong shadows, glare effects and high contrast intensity variations. Additionally, the acquisition room was installed with recessed ceiling-mounted LED lighting panels emitting a warm white to yellowish illumination so to maintain lighting conditions visually comparable to natural daylight. From each sieve fraction, a representative 75 g of soil sample was placed on a plain white paper. First, a set of three images were captured at predefined camera heights of 15 cm, 30 cm and 45 cm. Following this, the soil arrangement was manually disturbed using a paper folding and turning procedure. In particular, one corner of the paper was lifted and rotated inward in the opposite direction to overturn the top particle layer. The procedure was then repeated using other corners to ensure that previously underlying particles were brought to the surface while previously exposed particles were redistributed below. This disturbance procedure was intended to generate variability in particle exposure and arrangement. The same procedure was applied consistently across all samples and disturbance cycles. After each disturbance cycle, another set of three images were acquired at the predefined heights. The disturbance procedure was performed manually; therefore, no quantitative force or time measurement was imposed. Representative macroscopic and microscopic characterization of soil samples. (a) Shows smartphone images of soil captured at different heights (b) Shows microscope images of soil captured at different magnification. Cross-sectional measurements (indicated by red lines) reveal particle dimensions in the micrometer range. For each sieve fraction, 75 disturbance cycles were performed to introduce variability in particle arrangement, with 3 images recorded per cycle, resulting in 225 images per sieve fraction. In total, 1350 smartphone-based images were generated across all sieve fractions, forming a robust dataset for macroscopic analysis. Light intensity values were consistently monitored and recorded simultaneously during the image acquisition using Lightmeter iOS application. The recorded light intensity ranged from 74 to 861 lx. Lux measurements were recorded at three predefined heights of 15, 30 and 45 cm with the camera facing toward the soil sample during image acquisition Image parameters that may influence downstream image analysis were recorded for each image in associated Smartphone Image Parameters CSV file accompanying the dataset. These parameters include Focal length, aperture, shutter speed, height of camera, light intensity and image resolution. As the camera was operated with automatic exposure and autofocus settings, these parameters might differ per image; the EXIF metadata recorded with each image file indicates the actual settings applied and thus can be used to determine acquisition variability or to support normalization during preprocessing. The distance between the light source and soil sample was governed by the fixed room lighting configuration and window placement and no dedicated artificial illumination setup with controlled source to sample spacing was deployed. Lighting conditions were not kept consistent; ambient light levels fluctuated between 74 and 861 lx during image capture. These variations were documented for each image in the metadata, but because a standardized imaging environment was not used, differences in illumination contribute to uncontrolled variability throughout the dataset. Illumination levels were measured manually with a mobile light meter application at the time of picture taking. The data was then organized into a CSV metadata file for the purpose of providing reference information on the lighting conditions at the time of photo shooting The images captured were stored in JPEG format using the native camera output settings of the device. The smartphone camera was operated using the default automatic exposure and autofocus setting throughout the image acquisition process. No additional image compression or post-acquisition image enhancement procedure were applied, and images were archived and uploaded to the repository in their original acquisition format. No calibration target or colour reference was used for illumination normalization across smartphone acquisition procedure. The particle size metrics recorded in the dataset corresponds to the sieve fractions obtained by the mechanical sieve analysis procedure performed according to the ASTM/ISO guidelines. The Sieve Size (mm) column mentioned in the csv file in dataset uploaded on IEEE Dataport identifies the nominal particle size fraction associated with each image sample. The Table 5 describing all image parameter columns, associated units, and variable definitions for smartphone datasets is listed below. Table 5Description of parameter variables associated with smartphone-based soil particle images.Table 5: dummy alt textColumn NameDescriptionImage NameUnique file name corresponding to the smartphone imageSieve Size (mm)Particle size fraction collected from sieve analysis corresponding to the image sampleFocal Length (mm)Actual camera focal length extracted from the Image EXIF metadataAperture(f-number)Camera aperture value extracted from the Image MetadataShutter Speed (seconds)Exposure duration (camera shutter time) extracted from the Image MetadataHeight of Camera (cm)Vertical distance between smartphone camera and the soil sampleLight Intensity (lux)Ambient illumination level recorded during the image acquisition. An iOS application ‘Lightmeter’ was used to record the values. Range recorded: 74 to 861Image ResolutionSmartphone image resolution (width x height in pixels) Description of parameter variables associated with smartphone-based soil particle images. Microscopic image acquisition Microscopic image acquisition was performed for detailed particle-level analysis. Sub-samples were collected from five distinct locations within each sieve fraction and stored in Eppendorf tubes to preserve sample integrity. These sub-samples were recorded using an optical inverted microscope at varied magnifications based on particle size: Sample size of 1.18 mm and 0.6 mm at 4x (200 µm scale), sample size 0.3 mm at 10x (100 µm scale), and Sample sizes of 0.15 mm and 0.075 mm at 20x (50 µm scale). The microscope system was operated using the Olympus CKX53 inverted microscope equipped with a DP28 colour camera and Olympus cellSens Entry software. Each sub-sample generated 54 images, resulting in 270 images per fraction and a total of 1350 microscopic images the representative images are shown in Fig. 5.b. This multi-magnification strategy ensures a comprehensive visualization of particle shape, size distribution, and microstructural heterogeneity. The acquired images and corresponding information are provided in accompanying CSV file. The cross-sectional dimensions of representative particles were measured using the Olympus cellSens Entry software. For each image, the magnification was set up automatically depending on the objective lens used (4x, 10x and 20x, which corresponded to 200 µm, 100 µm and 50 µm). These measurements were directly rendered on the images as visible measurement lines with the corresponding length in micrometers. The measurement lines were placed manually and are therefore operator dependent. Table 6 below describes the parameter columns, associated units and variable definitions recorded for the microscopic image dataset. The metadata files deposited in the repository use the same column names as listed in Table 5, Table 6. Table 6Summary of image parameters and imaging conditions recorded for microscopic soil image acquisition.Table 6: dummy alt textColumn NameDescriptionSampleMicroscope sub-sample identifier. Example: 1.18 A - 1.18 E, 0.6 A - 0.6 E etc. Numeric values correspond to sieve fractions and letters denote independently collected sub-samples.Image NameUnique file name corresponding to the microscope imageMean IntensityCellSens software-reported mean intensity calculated from all image pixels over an 8-bit grayscale intensity range (0–255) and representing the overall image brightness. Range recorded: 88 to 241Height of Camera (mm)Working distance between microscope camera and glass slideMagnificationMagnification such as 4X, 10X, 20X for different sieve size particlesSieve Size (mm)Particle size fraction collected from sieve analysis corresponding to the image samplePixel CountAll microscopic images have a pixel count of 8922,096 pixels. Summary of image parameters and imaging conditions recorded for microscopic soil image acquisition. The reference data provided here comprise the sieve-fraction labels, the sieve-based particle size distribution and the magnification and corresponding calibrated scale recorded with each microscopic image. The rendered dimensional measurements represent a dimensional reference rather than particle level annotation, allowing a user to establish a known scale within the context of an external analysis software, thus enabling the morphometric descriptors for particles of interest. The measured values are not tabulated in the metadata file because the soil particles are irregular in shape and no single dimension can be defined that could represent the particle size across the dataset accurately. Table 7. Table 7Repository Manifest of the uploaded dataset.Table 7: dummy alt textRepository ComponentContentsFile FormatApproximate SizeSmartphone Image Dataset6 Folders containing 1350 images in totalJPEG2.67 GBMicroscope Image Dataset5 Folders with 25 Subfolders containing 1350 images in totalTIFF33.7 GBSmartphone Image Parameters1 FileCSV88 KBMicroscopic Image Parameters1 FileCSV69 KBREADME1 FileTXT6 KBLicense CC BY 4.01 FileTXT19 KB Repository Manifest of the uploaded dataset. Data processing compatibility The dataset supports the image processing and machine learning pipelines developed using Python and MATLAB facilitating feature extraction, segmentation and predictive modelling. The dataset deposited in IEEE DataPort is distributed under Creative Commons Attribution (CC BY) license in accordance with the repository policies. The users may access, reuse, redistribute and adapt the dataset provided that appropriate attribution is given to the original authors. Limitations Although illumination was maintained during smartphone image acquisition, the lack of standardized imaging chambers may introduce minor variability in illumination across samples. Additionally, automatic exposure and autofocus settings were not controlled or kept constant during image capture and therefore should be considered as potential confounding variables. Since the dataset is limited to only a single soil type, its applicability for generalized modeling across varied soil classes is constrained. Microscopic imaging was restricted to finer particle fractions, as the 2.36 mm fraction could not be adequately captured within the limited field of view at higher magnification levels of the microscope. Consequently, the microscopic image dataset spans five fractions (1.18 mm to 0.075 mm) compared with six for smartphone image dataset. The moisture content of the soil was not determined before sieve analysis. Residual moisture may influence particle aggregation, sieving efficiency and image appearance; therefore, this factor should be considered when interpreting the dataset. The storage history of the laboratory stock was also not documented. The dataset does not contain ground-truth segmentation masks or particle level annotations. For supervised segmentation and instance level morphometrics, ground truth must be obtained in an independent process, while for unsupervised algorithms such as thresholding, edge detection or morphological operations, the images can be processed directly. The dimensional measurement overlays are rendered into microscopic images and thus are part of the pixel data. Users performing segmentation, edge detection or intensity-based feature extraction should be aware that such overlay elements can be picked up as image features and may need to be masked or excluded in such analyses Ethics Statement The authors declare that the present study does not involve human subjects, animal experimentation or data derived from social media sources. Credit Author Statement Dr. Gobinath Ravindran: Conceptualization, Methodology, Software, Validation. Deep Arvinder: Investigation, Data curation, Writing, Original draft preparation. Rutuja Gumathannavar: Writing- Reviewing and Editing.