Pre-processing Workflow Research Articles

On the correlation between pre-processing workflow and dimensional accuracy of 3D printed parts in high-precision Material Jetting

Material Jetting (MJ) is distinguished in the additive manufacturing field for its ability to create accurate multi-material functional objects with features in the range of few micrometers. Analyzing the impact of each step in MJ from virtual design to 3D printed object on the process response is essential for optimizing the technology’s capabilities. The current state-of-the-art reveals a gap in MJ regarding the fundamental understanding of the correlation between the pre-processing workflow and the geometrical and dimensional accuracy. This study aims to bridge this gap by examining factors such as utilization of different CAD software with diverse working principles, part-based versus assembly-based design, alternative 3D model formats (STL, OBJ, 3MF, AMF, STP), and slicing approaches (using open-source slicers). Two test specimens, each containing ten elements (either cylinders or hemispheres) with diameters ranging from 254 µm to 12.7 mm are examined. The results demonstrate the significant influence of the aforementioned factors on geometrical and dimensional accuracy, except for 3D model formats. Specifically, for large elements, the achievable accuracy depends on the tessellation approach of the CAD system, while for very small elements, the rasterization process is defining, and designing in assembly mode can further enhance accuracy. The handling of the 3D model by the slicer particularly affects geometrical accuracy. Experimental validation confirms the impact of the pre-processing workflow, though the data-related dimensional deviation for the base layer is masked by the wetting behavior of the dispensed droplets on the substrate.

Comparative analysis of convolutional neural network and DenseNet121 transfer learning in agriculture focusing on crop leaf disease identification

PurposePlant cultivation holds a pivotal role in agriculture, necessitating precise disease identification for the overall health of plants. This research conducts a comprehensive comparative analysis between two prominent deep learning algorithms, convolutional neural network (CNN) and DenseNet121, with the goal of enhancing disease identification in tomato plant leaves.Design/methodology/approachThe dataset employed in this investigation is a fusion of primary data and publicly available data, covering 13 distinct disease labels and a total of 18,815 images for model training. The data pre-processing workflow prioritized activities such as normalizing pixel dimensions, implementing data augmentation and achieving dataset balance, which were subsequently followed by the modeling and testing phases.FindingsExperimental findings elucidated the superior performance of the DenseNet121 model over the CNN model in disease classification on tomato leaves. The DenseNet121 model attained a training accuracy of 98.27%, a validation accuracy of 87.47% and average recall, precision and F1-score metrics of 87, 88 and 87%, respectively. The ultimate aim was to implement the optimal classifier for a mobile application, namely Tanamin.id, and, therefore, DenseNet121 was the preferred choice.Originality/valueThe integration of private and public data significantly contributes to determining the optimal method. The CNN method achieves a training accuracy of 90.41% and a validation accuracy of 83.33%, whereas the DenseNet121 method excels with a training accuracy of 98.27% and a validation accuracy of 87.47%. The DenseNet121 architecture, comprising 121 layers, a global average pooling (GAP) layer and a dropout layer, showcases its effectiveness. Leveraging categorical_crossentropy as the loss function and utilizing the stochastic gradien descent (SGD) Optimizer with a learning rate of 0.001 guides the course of the training process. The experimental results unequivocally demonstrate the superior performance of DenseNet121 over CNN.

Improved hypertensive stroke classification based on multi-scale feature fusion of head axial CT angiogram and multimodal learning

Purpose:Strokes are severe cardiovascular and circulatory diseases with two main types: ischemic and hemorrhagic. Clinically, brain images such as computed tomography (CT) and computed tomography angiography (CTA) are widely used to recognize stroke types. However, few studies have combined imaging and clinical data to classify stroke or consider a factor as an Independent etiology. Methods:In this work, we propose a classification model that automatically distinguishes stroke types with hypertension as an independent etiology based on brain imaging and clinical data. We first present a preprocessing workflow for head axial CT angiograms, including noise reduction and feature enhancement of the images, followed by an extraction of regions of interest. Next, we develop a multi-scale feature fusion model that combines the location information of position features and the semantic information of deep features. Furthermore, we integrate brain imaging with clinical information through a multimodal learning model to achieve more reliable results. Results:Experimental results show our proposed models outperform state-of-the-art models on real imaging and clinical data, which reveals the potential of multimodal learning in brain disease diagnosis. Conclusion:The proposed methodologies can be extended to create AI-driven diagnostic assistance technology for categorizing strokes.

AUGMENTING THE PHILIPPINES’ DOST-ASTI’S POTENTIAL FLOOD EXTENTS MAPPING SERVICE WITH S-BAND NOVASAR-1 IMAGES

Abstract. The Philippines’ Advanced Science and Technology Institute under the Department of Science and Technology (DOST-ASTI) has developed an AI-based and near real-time flood extent mapping service that utilizes C-Band Sentinel-1 SAR images. However, this method is limited by the availability of the Sentinel-1 images during flooding events. To address this issue, the institute, through its SARwAIS Project, utilized the S-Band NovaSAR-1 satellite, which was designed and launched by Surrey Satellite Technology, Ltd. With a 10% share to NovaSAR-1’s imaging capacity, the country can task image acquisitions that could help augment the Sentinel-1 datasets. Successfully captured images are prepared using the institute’s developed pre-processing workflow. Afterwards, a thresholding method, adopted from UN-SPIDER’s recommended practices for flood mapping, is employed to identify potentially flooded areas from these images. Generated products are then assessed to determine their relative accuracy in detecting potential floods. Satisfactory products are then distributed to relevant disaster management agencies and are also published in the agency’s social media page for further information dissemination. Python scripts were then developed to automate the established workflows, which were initially done manually. These scripts also help expedite the generation of flood maps especially when processing multiple SAR images. The acquisition and utilization of NovaSAR-1 images substantially help the country address the gaps on the availability of workable data for a more informative disaster response especially during flooding events.

InferDB: In-Database Machine Learning Inference Using Indexes

The performance of inference with machine learning (ML) models and its integration with analytical query processing have become critical bottlenecks for data analysis in many organizations. An ML inference pipeline typically consists of a preprocessing workflow followed by prediction with an ML model. Current approaches for in-database inference implement preprocessing operators and ML algorithms in the database either natively, by transpiling code to SQL, or by executing user-defined functions in guest languages such as Python. In this work, we present a radically different approach that approximates an end-to-end inference pipeline (preprocessing plus prediction) using a light-weight embedding that discretizes a carefully selected subset of the input features and an index that maps data points in the embedding space to aggregated predictions of an ML model. We replace a complex preprocessing workflow and model-based inference with a simple feature transformation and an index lookup. Our framework improves inference latency by several orders of magnitude while maintaining similar prediction accuracy compared to the pipeline it approximates.

CoastalWQL: An Open-Source Tool for Drone-Based Mapping of Coastal Turbidity Using Push Broom Hyperspectral Imagery

Uncrewed-Aerial Vehicles (UAVs) and hyperspectral sensors are emerging as effective alternatives for monitoring water quality on-demand. However, image mosaicking for largely featureless coastal water surfaces or open seas has shown to be challenging. Another pertinent issue observed is the systematic image misalignment between adjacent flight lines due to the time delay between the UAV-borne sensor and the GNSS system. To overcome these challenges, this study introduces a workflow that entails a GPS-based image mosaicking method for push-broom hyperspectral images, together with a correction method to address the aforementioned systematic image misalignment. An open-source toolkit, CoastalWQL, was developed to facilitate the workflow, which includes essential pre-processing procedures for improving the image mosaic’s quality, such as radiometric correction, de-striping, sun glint correction, and object masking classification. For validation, UAV-based push-broom hyperspectral imaging surveys were conducted to monitor coastal turbidity in Singapore, and the implementation of CoastalWQL’s pre-processing workflow was evaluated at each step via turbidity retrieval. Overall, the results confirm that the image mosaicking of the push-broom hyperspectral imagery over featureless water surface using CoastalWQL with time delay correction enabled better localisation of the turbidity plume. Radiometric correction and de-striping were also found to be the most important pre-processing procedures, which improved turbidity prediction by 46.5%.

Machine learning in Huntington’s disease: exploring the Enroll-HD dataset for prognosis and driving capability prediction

BackgroundIn biomedicine, machine learning (ML) has proven beneficial for the prognosis and diagnosis of different diseases, including cancer and neurodegenerative disorders. For rare diseases, however, the requirement for large datasets often prevents this approach. Huntington’s disease (HD) is a rare neurodegenerative disorder caused by a CAG repeat expansion in the coding region of the huntingtin gene. The world’s largest observational study for HD, Enroll-HD, describes over 21,000 participants. As such, Enroll-HD is amenable to ML methods. In this study, we pre-processed and imputed Enroll-HD with ML methods to maximise the inclusion of participants and variables. With this dataset we developed models to improve the prediction of the age at onset (AAO) and compared it to the well-established Langbehn formula. In addition, we used recurrent neural networks (RNNs) to demonstrate the utility of ML methods for longitudinal datasets, assessing driving capabilities by learning from previous participant assessments.ResultsSimple pre-processing imputed around 42% of missing values in Enroll-HD. Also, 167 variables were retained as a result of imputing with ML. We found that multiple ML models were able to outperform the Langbehn formula. The best ML model (light gradient boosting machine) improved the prognosis of AAO compared to the Langbehn formula by 9.2%, based on root mean squared error in the test set. In addition, our ML model provides more accurate prognosis for a wider CAG repeat range compared to the Langbehn formula. Driving capability was predicted with an accuracy of 85.2%. The resulting pre-processing workflow and code to train the ML models are available to be used for related HD predictions at: https://github.com/JasperO98/hdml/tree/main.ConclusionsOur pre-processing workflow made it possible to resolve the missing values and include most participants and variables in Enroll-HD. We show the added value of a ML approach, which improved AAO predictions and allowed for the development of an advisory model that can assist clinicians and participants in estimating future driving capability.

Utilising Sentinel-1's Orbital Stability for Efficient Pre-Processing of Radiometric Terrain Corrected Gamma Nought Backscatter.

Radiometric Terrain Corrected (RTC) gamma nought backscatter, which was introduced around a decade ago, has evolved into the standard for analysis-ready Synthetic Aperture Radar (SAR) data. While working with RTC backscatter data is particularly advantageous over undulated terrain, it requires substantial computing resources given that the terrain flattening is more computationally demanding than simple orthorectification. The extra computation may become problematic when working with large SAR datasets such as the one provided by the Sentinel-1 mission. In this study, we examine existing Sentinel-1 RTC pre-processing workflows and assess ways to reduce processing and storage overheads by considering the satellite's high orbital stability. By propagating Sentinel-1's orbital deviations through the complete pre-processing chain, we show that the local contributing area and the shadow mask can be assumed to be static for each relative orbit. Providing them as a combined external static layer to the pre-processing workflow, and streamlining the transformations between ground and orbit geometry, reduces the overall processing times by half. We conducted our experiments with our in-house developed toolbox named wizsard, which allowed us to analyse various aspects of RTC, specifically run time performance, oversampling, and radiometric quality. Compared to the Sentinel Application Platform (SNAP) this implementation allowed speeding up processing by factors of 10-50. The findings of this study are not just relevant for Sentinel-1 but for all SAR missions with high spatio-temporal coverage and orbital stability.

Successful reproduction of a large EEG study across software packages

As an active field of research and with the development of state-of-the-art algorithms to analyze EEG datasets, the parametrization of Electroencephalography (EEG) analysis workflows has become increasingly flexible and complex, with a great variety of methodological options and tools to be selected at each step. This high analytical flexibility can be problematic as it can yield to variability in research outcomes. Therefore, growing attention has been recently paid to understand the potential impact of different methodological decisions on the reproducibility of results.In this paper, we aim to examine how sensitive the results of EEG analyses are to variations in preprocessing with different software tools. We reanalyzed the shared EEG data (N = 500) from (Williams et al., 2021) using three of the most commonly used open-source Matlab-based EEG software tools: EEGLAB, Brainstorm and FieldTrip. After reproducing the same original preprocessing workflow in each software, the resulting event-related potentials (ERPs) were qualitatively and quantitatively compared in order to examine the degree of consistency/discrepancy between software packages. Our findings show a good degree of convergence in terms of the general profile of ERP waveforms, peak latencies and effect size estimates related to specific signal features. However, considerable variability was also observed in the magnitude of the absolute voltage observed with each software package as reflected by the similarity values and observed statistical differences at particular channels and time instants. In conclusion, we believe that this study provides valuable clues to better understand the impact of the software tool on the analysis of EEG results.

A deep learning framework for defect prediction based on thermographic in-situ monitoring in laser powder bed fusion

AbstractThe prediction of porosity is a crucial task for metal based additive manufacturing techniques such as laser powder bed fusion. Short wave infrared thermography as an in-situ monitoring tool enables the measurement of the surface radiosity during the laser exposure. Based on the thermogram data, the thermal history of the component can be reconstructed which is closely related to the resulting mechanical properties and to the formation of porosity in the part. In this study, we present a novel framework for the local prediction of porosity based on extracted features from thermogram data. The framework consists of a data pre-processing workflow and a supervised deep learning classifier architecture. The data pre-processing workflow generates samples from thermogram feature data by including feature information from multiple subsequent layers. Thereby, the prediction of the occurrence of complex process phenomena such as keyhole pores is enabled. A custom convolutional neural network model is used for classification. The model is trained and tested on a dataset from thermographic in-situ monitoring of the manufacturing of an AISI 316L stainless steel test component. The impact of the pre-processing parameters and the local void distribution on the classification performance is studied in detail. The presented model achieves an accuracy of 0.96 and an f1-Score of 0.86 for predicting keyhole porosity in small sub-volumes with a dimension of (700 × 700 × 50) µm3. Furthermore, we show that pre-processing parameters such as the porosity threshold for sample labeling and the number of included subsequent layers are influential for the model performance. Moreover, the model prediction is shown to be sensitive to local porosity changes although it is trained on binary labeled data that disregards the actual sample porosity.

Satellite-Derived Bathymetry for Selected Shallow Maltese Coastal Zones

Bathymetric information has become essential to help maintain and operate coastal zones. Traditional in situ bathymetry mapping using echo sounders is inefficient in shallow waters and operates at a high logistical cost. On the other hand, lidar mapping provides an efficient means of mapping coastal areas. However, this comes at a high acquisition cost as well. In comparison, satellite-derived bathymetry (SDB) provides a more cost-effective way of mapping coastal regions, albeit at a lower resolution. This work utilises all three of these methods collectively, to obtain accurate bathymetric depth data of two pocket beaches, Golden Bay and Għajn Tuffieħa, located in the northwestern region of Malta. Using the Google Earth Engine platform, together with Sentinel-2 data and collected in situ measurements, an empirical pre-processing workflow for estimating SDB was developed. Four different machine learning algorithms which produced differing depth accuracies by calibrating SDBs with those derived from alternative techniques were tested. Thus, this study provides an insight into the depth accuracy that can be achieved for shallow coastal regions using SDB techniques.

FMRIflows: A Consortium of Fully Automatic Univariate and Multivariate fMRI Processing Pipelines

How functional magnetic resonance imaging (fMRI) data are analyzed depends on the researcher and the toolbox used. It is not uncommon that the processing pipeline is rewritten for each new dataset. Consequently, code transparency, quality control and objective analysis pipelines are important for improving reproducibility in neuroimaging studies. Toolboxes, such as Nipype and fMRIPrep, have documented the need for and interest in automated pre-processing analysis pipelines. Recent developments in data-driven models combined with high resolution neuroimaging dataset have strengthened the need not only for a standardized preprocessing workflow, but also for a reliable and comparable statistical pipeline. Here, we introduce fMRIflows: a consortium of fully automatic neuroimaging pipelines for fMRI analysis, which performs standard preprocessing, as well as 1st- and 2nd-level univariate and multivariate analyses. In addition to the standardized pre-processing pipelines, fMRIflows provides flexible temporal and spatial filtering to account for datasets with increasingly high temporal resolution and to help appropriately prepare data for advanced machine learning analyses, improving signal decoding accuracy and reliability. This paper first describes fMRIflows’ structure and functionality, then explains its infrastructure and access, and lastly validates the toolbox by comparing it to other neuroimaging processing pipelines such as fMRIPrep, FSL and SPM. This validation was performed on three datasets with varying temporal sampling and acquisition parameters to prove its flexibility and robustness. fMRIflows is a fully automatic fMRI processing pipeline which uniquely offers univariate and multivariate single-subject and group analyses as well as pre-processing.

Measuring proteins in H2O using 2D-IR spectroscopy: pre-processing steps and applications toward a protein library.

The ability of two-dimensional infrared (2D-IR) spectroscopy to measure the amide I band of proteins in H2O rather than D2O-based solvents by evading the interfering water signals has enabled in vivo studies of proteins under physiological conditions and in biofluids. Future exploitation of 2D-IR in analytical settings, from diagnostics to protein screening, will, however, require comparisons between multiple datasets, necessitating control of data collection protocols to minimize measurement-to-measurement inconsistencies. Inspired by analytical spectroscopy applications in other disciplines, we describe a workflow for pre-processing 2D-IR data that aims to simplify spectral cross-comparisons. Our approach exploits the thermal water signal that is collected simultaneously with, but is temporally separated from the amide I response to guide custom baseline correction and spectral normalization strategies before combining them with Principal Component noise reduction tools. Case studies show that application of elements of the pre-processing workflow to previously published data enables improvements in quantification accuracy and detection limits. We subsequently apply the complete workflow in a new pilot study, testing the ability of a prototype library of 2D-IR spectra to quantify the four major protein constituents of blood serum in a single, label-free measurement. These advances show progress toward the robust data handling strategies that will be necessary for future applications of 2D-IR to pharmaceutical or biomedical problems.

Comparative meta-analysis of host transcriptional response during Streptococcus pneumoniae carriage or infection

Infection with Streptococcus pneumoniae causes over 150,000 cases of pneumonia annually in the United States alone. We performed a meta-analysis of publicly available RNA-sequencing data to compare the host-specific intracellular transcriptional responses that differ between infection and carriage with S. pneumoniae in humans and mice. We applied an automated preprocessing workflow to collections of publicly available fastq files that were categorized as either of two phenotypes—infection or carriage in humans and mice. We identified the differentially expressed genes and intracellular signaling pathways that changed in host cells during infection or carriage and whether these human phenotypes could be appropriately modeled at the molecular level in mice. Although we observed multiple differentially expressed genes shared among multiple comparisons, we found no overlapping significant signaling pathways between the mouse and human studies in either phenotype. Based on the samples included in this secondary analysis, we identified a minimal number of generalized transcriptional relationships between host infection and carriage phenotypes of S. pneumoniae that were consistently shared between the mouse and human hosts. Our findings suggest that additional controlled datasets in mouse and human carriage or infection models are needed to better understand the underlying mechanism(s) of invasion and pathogenesis. This knowledge could then contribute to the development of improved prophylactics and/or therapeutics against this pathogen.

Abnormality classification from electrocardiograms with various lead combinations

Objective. As cardiovascular diseases are a leading cause of death, early and accurate diagnosis of cardiac abnormalities for a lower cost becomes particularly important. Given electrocardiogram (ECG) datasets from multiple sources, there exist many challenges to the development of generalized models that can identify multiple types of cardiac abnormalities from both 12-lead ECG signals and reduced-lead ECG signals. In this study, our objective is to build robust models that can accurately classify 30 types of abnormalities from various lead combinations of ECG signals. Approach. Given the challenges of this problem, we propose a framework for building robust models for ECG signal classification. Firstly, a preprocessing workflow is adopted for each ECG dataset to mitigate the problem of data divergence. Secondly, to capture the lead-wise relations, we use a squeeze-and-excitation deep residual network as our base model. Thirdly, we propose a cross-relabeling strategy and apply the sign-augmented loss function to tackle the corrupted labels in the data. Furthermore, we utilize a pos-if-any-pos ensemble strategy and a dataset-wise cross-evaluation strategy to handle the uncertainty of the data distribution in the application. Main results. In the Physionet/Computing in Cardiology Challenge 2021, our approach achieved the challenge metric scores of 0.57, 0.59, 0.59, 0.58, 0.57 on 12-, 6-, 4-, 3- and 2-lead versions and an averaged challenge metric score of 0.58 over all the lead versions. Significance. Using the proposed framework, we have developed the models from several large datasets with sufficiently labeled abnormalities. Our models are able to identify 30 ECG abnormalities accurately based on various lead combinations of ECG signals. The performance on hidden test data demonstrates the effectiveness of the proposed approaches.

Three-dimensional reflection seismic imaging of the iron oxide deposits in the Ludvika mining area, Sweden, using Fresnel volume migration

Abstract. We present pre-stack depth-imaging results for a case study of 3D reflection seismic exploration at the Blötberget iron oxide mining site belonging to the Bergslagen mineral district in central Sweden. The goal of the study is to directly image the ore-bearing horizons and to delineate their possible depth extension below depths known from existing boreholes. For this purpose, we applied a tailored pre-processing workflow and two different seismic imaging approaches, Kirchhoff pre-stack depth migration (KPSDM) and Fresnel volume migration (FVM). Both imaging techniques deliver a well-resolved 3D image of the deposit and its host rock, where the FVM image yields a significantly better image quality compared to the KPSDM image. We were able to unravel distinct horizons, which are linked to known mineralization and provide insights on their possible lateral and depth extent. Comparison of the known mineralization with the final FVM reflection volume suggests a good agreement of the position and the shape of the imaged reflectors caused by the mineralization. Furthermore, the images show additional reflectors below the mineralization and reflectors with opposite dips. One of these reflectors is interpreted to be a fault intersecting the mineralization, which can be traced to the surface and linked to a fault trace in the geological map. The depth-imaging results can serve as the basis for further investigations, drilling, and follow-up mine planning at the Blötberget mining site..

Enhancing body waves in passive seismic reflection exploration: A case study in Inner Mongolia, China

Among all geophysical exploration methods, seismic exploration is undoubtedly the most important due to its ability to allow depth exploration at high resolutions. Traditionally speaking, the method needs an active seismic source, such as dynamite, to generate energy and perform reflection and refractions. An active source usually means high cost, and it also can be quite difficult to implement when surface conditions are particularly complex. The use of passive seismic for reflection exploration does not require an active seismic source. It has the potential of providing a low-cost alternative technique in some exploration areas. The main issues related to passive-source body-wave exploration include suppressing the surface waves retrieved from sources located at or near the surface, as well as enhancing the body waves from random sources at depth. We address these problems by developing a preprocessing workflow to suppress the surface waves and the other unwanted coherent noise events in the original data without seriously affecting the remaining body waves. We also propose a method for separating surface and body waves based on the signal-to-noise ratio of the frequency-domain signals. Next, we use crosscorrelation to generate virtual shot gathers, just like the active ones, and use conventional seismic data processing steps to generate the final seismic imaging. By analyzing the passive data set collected from Inner Mongolia, we have verified the applicability of the proposed method, and the retrieved final stack section indicates good consistency with an active seismic stack section along the same line. Accordingly, we assert that the application of this data processing method will contribute to the body-wave imaging and the inversion analysis of passive seismic records.

A systematic review and meta-analysis of Digital elevation model (DEM) fusion: pre-processing, methods and applications

The remote sensing community has identified data fusion as one of the key challenging topics of the 21st century. The subject of image fusion in two-dimensional (2D) space has been covered in several published reviews. However, the special case of 2.5D/3D Digital Elevation Model (DEM) fusion has not been addressed till date. DEM fusion is a key application of data fusion in remote sensing. It takes advantage of the complementary characteristics of multi-source DEMs to deliver a more complete, accurate, and reliable elevation dataset. Although several methods for fusing DEMs have been developed, the absence of a well-rounded review has limited their proliferation among researchers and end-users. Combining knowledge from multiple studies is often required to inform a holistic perspective and guide further research. In response, this paper provides a systematic review of DEM fusion: the pre-processing workflow, methods and applications, enhanced with a meta-analysis. Through the discussion and comparative analysis, unresolved challenges and open issues are identified, and future directions for research are proposed. This review is a timely solution and an invaluable source of information for researchers within the fields of remote sensing and spatial information science, and the data fusion community at large.

Travel-mode classification based on GPS-trajectory data and geographic information using an XGBoost classifier

Massive Global Positioning System (GPS) trajectory datasets are being produced owing to the advances in mobile sensors, the Internet, and GPS devices. Accurately inferring travel modes from GPS trajectory data can be helpful in transportation planning and modeling, infrastructure design, etc. However, adverse factors such as data noise, differences in sampling rate, and inadequate features have a negative impact on the results of travel mode classification. In this paper, to address such issues, we first propose a preprocessing workflow, which includes data cleaning, segmentation, and resampling, to preprocess raw trajectories. Then, we add new features related to the road and bus stop information for travel mode classification using an XGBoost (eXtreme Gradient Boosting) classifier, along with various basic features of the trajectories. We conducted a set of experiments on the GeoLife dataset using a group of state-of-the-art methods. The results showed that the proposed methods can improve the classification accuracy by using all the classifiers we compared and the classification accuracy using the XGBoost classifier can reach a maximum of 90.41%.

Computer vision to recognize construction waste compositions: A novel boundary-aware transformer (BAT) model

Recognition of construction waste compositions using computer vision (CV) is increasingly explored to enable its subsequent management, e.g., determining chargeable levy at disposal facilities or waste sorting using robot arms. However, the applicability of existing CV-enabled construction waste recognition in real-life scenarios is limited by their relatively low accuracy, characterized by a failure to distinguish boundaries among different waste materials. This paper aims to propose a novel boundary-aware Transformer (BAT) model for fine-grained composition recognition of construction waste mixtures. First, a pre-processing workflow is devised to separate the hard-to-recognize edges from the background. Second, a Transformer structure with a self-designed cascade decoder is developed to segment different waste materials from construction waste mixtures. Finally, a learning-enabled edge refinement scheme is used to fine-tune the ignored boundaries, further boosting the segmentation precision. The performance of the BAT model was evaluated on a benchmark dataset comprising nine types of materials in a cluttered and mixture state. It recorded a 5.48% improvement of MIoU (mean intersection over union) and 3.65% of MAcc (Mean Accuracy) against the baseline. The research contributes to the body of interdisciplinary knowledge by presenting a novel deep learning model for construction waste material semantic segmentation. It can also expedite the applications of CV in construction waste management to achieve a circular economy.