Large Image Datasets Research Articles

Traffic Sign Classification Using ASIC

Abstract: The main aim of this project is to focus on traffic signs classification using Tensor flow and deploying the model onto an ASIC for engineering applications. The classification is done using Convolutional Neural Networks (CNNs) in TensorFlow, where the model is trained using a large dataset of traffic sign images. The training images are pre-processed and labelled, and the model is fine-tuned using transfer learning techniques. After the model is trained, it is tested with a separate dataset to ensure its accuracy. Once the model is finalized, it is optimized for deployment onto an ASIC using Micro Python. The ASIC is a custom-designed hardware that is optimized for the specific task of traffic sign classification. The performance of the ASIC is then compared with a standard computer to evaluate its speed and efficiency. This project demonstrates the ability to implement machine learning algorithms onto custom-designed hardware for specific engineering applications, specifically for the purpose of traffic sign classification.

Comparison of deep learning-based image segmentation methods for intravascular ultrasound on retrospective and large image cohort study

ObjectivesThe aim of this study was to investigate the generalization performance of deep learning segmentation models on a large cohort intravascular ultrasound (IVUS) image dataset over the lumen and external elastic membrane (EEM), and to assess the consistency and accuracy of automated IVUS quantitative measurement parameters.MethodsA total of 11,070 IVUS images from 113 patients and pullbacks were collected and annotated by cardiologists to train and test deep learning segmentation models. A comparison of five state of the art medical image segmentation models was performed by evaluating the segmentation of the lumen and EEM. Dice similarity coefficient (DSC), intersection over union (IoU) and Hausdorff distance (HD) were calculated for the overall and for subsets of different IVUS image categories. Further, the agreement between the IVUS quantitative measurement parameters calculated by automatic segmentation and those calculated by manual segmentation was evaluated. Finally, the segmentation performance of our model was also compared with previous studies.ResultsCENet achieved the best performance in DSC (0.958 for lumen, 0.921 for EEM) and IoU (0.975 for lumen, 0.951 for EEM) among all models, while Res-UNet was the best performer in HD (0.219 for lumen, 0.178 for EEM). The mean intraclass correlation coefficient (ICC) and Bland–Altman plot demonstrated the extremely strong agreement (0.855, 95% CI 0.822–0.887) between model's automatic prediction and manual measurements.ConclusionsDeep learning models based on large cohort image datasets were capable of achieving state of the art (SOTA) results in lumen and EEM segmentation. It can be used for IVUS clinical evaluation and achieve excellent agreement with clinicians on quantitative parameter measurements.

Self-supervised pre-training with contrastive and masked autoencoder methods for dealing with small datasets in deep learning for medical imaging

Deep learning in medical imaging has the potential to minimize the risk of diagnostic errors, reduce radiologist workload, and accelerate diagnosis. Training such deep learning models requires large and accurate datasets, with annotations for all training samples. However, in the medical imaging domain, annotated datasets for specific tasks are often small due to the high complexity of annotations, limited access, or the rarity of diseases. To address this challenge, deep learning models can be pre-trained on large image datasets without annotations using methods from the field of self-supervised learning. After pre-training, small annotated datasets are sufficient to fine-tune the models for a specific task. The most popular self-supervised pre-training approaches in medical imaging are based on contrastive learning. However, recent studies in natural image processing indicate a strong potential for masked autoencoder approaches. Our work compares state-of-the-art contrastive learning methods with the recently introduced masked autoencoder approach “SparK” for convolutional neural networks (CNNs) on medical images. Therefore, we pre-train on a large unannotated CT image dataset and fine-tune on several CT classification tasks. Due to the challenge of obtaining sufficient annotated training data in medical imaging, it is of particular interest to evaluate how the self-supervised pre-training methods perform when fine-tuning on small datasets. By experimenting with gradually reducing the training dataset size for fine-tuning, we find that the reduction has different effects depending on the type of pre-training chosen. The SparK pre-training method is more robust to the training dataset size than the contrastive methods. Based on our results, we propose the SparK pre-training for medical imaging tasks with only small annotated datasets.

An efficient semi-supervised quality control system trained using physics-based MRI-artefact generators and adversarial training

Large medical imaging data sets are becoming increasingly available. A common challenge in these data sets is to ensure that each sample meets minimum quality requirements devoid of significant artefacts. Despite a wide range of existing automatic methods having been developed to identify imperfections and artefacts in medical imaging, they mostly rely on data-hungry methods. In particular, the scarcity of artefact-containing scans available for training has been a major obstacle in the development and implementation of machine learning in clinical research. To tackle this problem, we propose a novel framework having four main components: (1) a set of artefact generators inspired by magnetic resonance physics to corrupt brain MRI scans and augment a training dataset, (2) a set of abstract and engineered features to represent images compactly, (3) a feature selection process that depends on the class of artefact to improve classification performance, and (4) a set of Support Vector Machine (SVM) classifiers trained to identify artefacts. Our novel contributions are threefold: first, we use the novel physics-based artefact generators to generate synthetic brain MRI scans with controlled artefacts as a data augmentation technique. This will avoid the labour-intensive collection and labelling process of scans with rare artefacts. Second, we propose a large pool of abstract and engineered image features developed to identify 9 different artefacts for structural MRI. Finally, we use an artefact-based feature selection block that, for each class of artefacts, finds the set of features that provide the best classification performance. We performed validation experiments on a large data set of scans with artificially-generated artefacts, and in a multiple sclerosis clinical trial where real artefacts were identified by experts, showing that the proposed pipeline outperforms traditional methods. In particular, our data augmentation increases performance by up to 12.5 percentage points on the accuracy, F1, F2, precision and recall. At the same time, the computation cost of our pipeline remains low –less than a second to process a single scan– with the potential for real-time deployment. Our artefact simulators obtained using adversarial learning enable the training of a quality control system for brain MRI that otherwise would have required a much larger number of scans in both supervised and unsupervised settings. We believe that systems for quality control will enable a wide range of high-throughput clinical applications based on the use of automatic image-processing pipelines.

Subsurface imaging dataset acquired at the Garner Valley Downhole Array site using a dense network of three-component nodal stations

There is a growing need to characterize the engineering material properties of the shallow subsurface in three dimensions for advanced engineering analyses. However, imaging the near-surface in three dimensions at spatial resolutions required for such purposes remains in its infancy and requires further study before it can be adopted into practice. To enable and accelerate research in this area, we present a large subsurface imaging data set acquired using a dense network of three-component (3C) nodal stations acquired in 2019 at the Garner Valley Downhole Array (GVDA) site. Acquisition of this data set involved the deployment of 196 stations positioned on a 14 × 14 grid with a 5 m spacing. The array was used to acquire active-source data generated by a vibroseis truck and an instrumented sledgehammer, and passive-wavefield data containing ambient noise. The active-source acquisition included 66 vibroseis and 209 instrumented sledgehammer source locations. Multiple source impacts were recorded at each source location to enable stacking of the recorded signals. The active-source recordings are provided in terms of both raw, uncorrected units of counts and corrected engineering units of meters per second. For each source impact, the force output from the vibroseis or instrumented sledgehammer was recorded and is provided in both raw counts and engineering units of kilonewtons. The passive-wavefield data include 28 h of ambient noise recorded over two nighttime deployments. The data set is shown to be useful for active-source and passive-wavefield three-dimensional imaging and other subsurface characterization techniques, which include horizontal-to-vertical spectral ratios (HVSRs), multichannel analysis of surface waves (MASW), and microtremor array measurements (MAM).

An Enhanced Automated Identification of Brain Tumor Cells Using Image Segmentation

Brain tumors are a serious threat to human health, and getting a proper diagnosis quickly is essential for successful treatment. This study introduces a novel automated method that uses improved image segmentation algorithms to detect brain tumor cells in diagnostic images. The work uses cutting-edge deep learning models and novel preprocessing techniques to boost the precision and speed with which tumor cells may be identified. The procedure starts with gathering a large dataset of brain tumor images, then moves on to intensive preprocessing to improve the quality of those images. To accurately define tumor locations, a new picture segmentation approach is developed that utilizes convolutional neural networks (CNNs) and morphological operations. The segmented sections are then used to develop a deep learning classifier to recognize tumor cells. The proposed method has been shown to be effective in experiments, with a noticeable increase in the accuracy of identifying tumor cells over previous methods. On a large test dataset, the system demonstrates clinical viability with an average classification using Average polling technique of 92% and classification using Max polling technique of 98.87%. In conclusion, this study advances medical image analysis by providing a more effective automated method of identifying brain tumor cells.

A computer vision and residual neural network (ResNet) combined method for automated and accurate yeast replicative aging analysis of high-throughput microfluidic single-cell images

With the rapid development of microfluidic platforms in high-throughput single-cell culturing, laborious operation to manipulate massive budding yeast cells (Saccharomyces cerevisiae) in replicative aging studies has been greatly simplified and automated. As a result, large datasets of microscopy images bring challenges to fast and accurately determine yeast replicative lifespan (RLS), which is the most important parameter to study cell aging. Based on our microfluidic diploid yeast long-term culturing (DYLC) chip that features 1100 traps to immobilize single cells and record their proliferation and aging via time-lapse imaging, herein, a dedicated algorithm combined with computer vision and residual neural network (ResNet) was presented to efficiently process tremendous micrographs in a high-throughput and automated manner. The image-processing algorithm includes following pivotal steps: (i) segmenting multi-trap micrographs into time-lapse single-trap sub-images, (ii) labeling 8 yeast budding features and training the 18-layer ResNet, (iii) converting the ResNet predictions in analog values into digital signals, (iv) recognizing cell dynamic events, and (v) determining yeast RLS and budding time interval (BTI) ultimately. The ResNet algorithm achieved high F1 scores (over 92%) demonstrating the effectiveness and accuracy in the recognition of yeast budding events, such as bud appearance, daughter dissection and cell death. Therefore, the results conduct that similar deep learning algorithms could be tailored to analyze high-throughput microscopy images and extract multiple cell behaviors in microfluidic single-cell analysis.

Updates in Diagnostic Imaging for Infectious Keratitis: A Review.

Infectious keratitis (IK) is among the top five leading causes of blindness globally. Early diagnosis is needed to guide appropriate therapy to avoid complications such as vision impairment and blindness. Slit lamp microscopy and culture of corneal scrapes are key to diagnosing IK. Slit lamp photography was transformed when digital cameras and smartphones were invented. The digital camera or smartphone camera sensor's resolution, the resolution of the slit lamp and the focal length of the smartphone camera system are key to a high-quality slit lamp image. Alternative diagnostic tools include imaging, such as optical coherence tomography (OCT) and in vivo confocal microscopy (IVCM). OCT's advantage is its ability to accurately determine the depth and extent of the corneal ulceration, infiltrates and haze, therefore characterizing the severity and progression of the infection. However, OCT is not a preferred choice in the diagnostic tool package for infectious keratitis. Rather, IVCM is a great aid in the diagnosis of fungal and Acanthamoeba keratitis with overall sensitivities of 66-74% and 80-100% and specificity of 78-100% and 84-100%, respectively. Recently, deep learning (DL) models have been shown to be promising aids for the diagnosis of IK via image recognition. Most of the studies that have developed DL models to diagnose the different types of IK have utilised slit lamp photographs. Some studies have used extremely efficient single convolutional neural network algorithms to train their models, and others used ensemble approaches with variable results. Limitations of DL models include the need for large image datasets to train the models, the difficulty in finding special features of the different types of IK, the imbalance of training models, the lack of image protocols and misclassification bias, which need to be overcome to apply these models into real-world settings. Newer artificial intelligence technology that generates synthetic data, such as generative adversarial networks, may assist in overcoming some of these limitations of CNN models.

Smart Plant Disease Analysis and Management System Using Ai and Iot

Plant diseases are one of the major causes of crop yield reduction worldwide. Early detection and effective management of plant diseases will ensure a sustainable food supply and reduce famine. In this project, we propose a Plant Disease Management System (PDMS) that uses Artificial Intelligence (AI) and Internet of Things (IoT) technologies to diagnose and manage plant diseases. The proposed system consists of two main components: a plant disease diagnosis module and a disease management module. The plant disease diagnosis module uses AI algorithms to analyze images of plant leaves and identify the type of disease affecting the plant. The AI model is trained on a large dataset of plant images using deep learning techniques, such as Convolution Neural Networks (CNNs) and Transfer Learning.

ANDA: an open-source tool for automated image analysis of in vitro neuronal cells

BackgroundImaging of in vitro neuronal differentiation and measurements of cell morphologies have led to novel insights into neuronal development. Live-cell imaging techniques and large datasets of images have increased the demand for automated pipelines for quantitative analysis of neuronal morphological metrics.ResultsANDA is an analysis workflow that quantifies various aspects of neuronal morphology from high-throughput live-cell imaging screens of in vitro neuronal cell types. This tool automates the analysis of neuronal cell numbers, neurite lengths and neurite attachment points. We used chicken, rat, mouse, and human in vitro models for neuronal differentiation and have demonstrated the accuracy, versatility, and efficiency of the tool.ConclusionsANDA is an open-source tool that is easy to use and capable of automated processing from time-course measurements of neuronal cells. The strength of this pipeline is the capability to analyse high-throughput imaging screens.

Ten recommendations for organising bioimaging data for archival

Organised data is easy to use but the growth of bioimaging, with improvements in instrumentation, detectors, software and experimental techniques has resulted in an explosion in the volumes of data being generated, making this an elusive goal. This guide offers a handful of recommendations whose implementation would contribute towards better organised data in preparation for archival. Based on our experience archiving large image datasets in EMPIAR, the BioImage Archive and BioStudies, we propose a number of strategies that we believe would make future data depositions more useful to the bioimaging community and that may also find use in other data-intensive disciplines. To facilitate the process of analysing data organisation, we present bandbox, a Python package that provides users with an assessment of their data by flagging potential issues that could be addressed before archival.

Chasing a Better Decision Margin for Discriminative Histopathological Breast Cancer Image Classification

When considering a large dataset of histopathologic breast images captured at various magnification levels, the process of distinguishing between benign and malignant cancer from these images can be time-intensive. The automation of histopathological breast cancer image classification holds significant promise for expediting pathology diagnoses and reducing the analysis time. Convolutional neural networks (CNNs) have recently gained traction for their ability to more accurately classify histopathological breast cancer images. CNNs excel at extracting distinctive features that emphasize semantic information. However, traditional CNNs employing the softmax loss function often struggle to achieve the necessary discriminatory power for this task. To address this challenge, a set of angular margin-based softmax loss functions have emerged, including angular softmax (A-Softmax), large margin cosine loss (CosFace), and additive angular margin (ArcFace), each sharing a common objective: maximizing inter-class variation while minimizing intra-class variation. This study delves into these three loss functions and their potential to extract distinguishing features while expanding the decision boundary between classes. Rigorous experimentation on a well-established histopathological breast cancer image dataset, BreakHis, has been conducted. As per the results, it is evident that CosFace focuses on augmenting the differences between classes, while A-Softmax and ArcFace tend to emphasize augmenting within-class variations. These observations underscore the efficacy of margin penalties on angular softmax losses in enhancing feature discrimination within the embedding space. These loss functions consistently outperform softmax-based techniques, either by widening the gaps among classes or enhancing the compactness of individual classes.

Retina Oculomics in Neurodegenerative Disease.

Ophthalmic biomarkers have long played a critical role in diagnosing and managing ocular diseases. Oculomics has emerged as a field that utilizes ocular imaging biomarkers to provide insights into systemic diseases. Advances in diagnostic and imaging technologies including electroretinography, optical coherence tomography (OCT), confocal scanning laser ophthalmoscopy, fluorescence lifetime imaging ophthalmoscopy, and OCT angiography have revolutionized the ability to understand systemic diseases and even detect them earlier than clinical manifestations for earlier intervention. With the advent of increasingly large ophthalmic imaging datasets, machine learning models can be integrated into these ocular imaging biomarkers to provide further insights and prognostic predictions of neurodegenerative disease. In this manuscript, we review the use of ophthalmic imaging to provide insights into neurodegenerative diseases including Alzheimer Disease, Parkinson Disease, Amyotrophic Lateral Sclerosis, and Huntington Disease. We discuss recent advances in ophthalmic technology including eye-tracking technology and integration of artificial intelligence techniques to further provide insights into these neurodegenerative diseases. Ultimately, oculomics opens the opportunity to detect and monitor systemic diseases at a higher acuity. Thus, earlier detection of systemic diseases may allow for timely intervention for improving the quality of life in patients with neurodegenerative disease.

MASIC: Deep Mask Stereo Image Compression

Stereo image compression (SIC) aims to simultaneously compress a pair of left and right stereoscopic images, which can achieve higher compression efficiency than single image compression. In this paper, to benefit the SIC tasks, we collect a large real-world stereo image dataset, namely Palace, which is composed of hundreds of stereo image pairs at high-resolution. More importantly, we propose a novel mask stereo image compression network, namely MASIC, which can jointly compress the stereo images with high compression efficiency. Specifically, we first estimate the homography matrix between the stereo images through a regression model. Then, the left image is spatially transformed by the homography matrix, so that only the residual information needs to be encoded for the right image. To avoid the wrong guidance between stereo image pair, we propose a mask prediction module (MPM) to generate a multi-channel guided mask to navigate both the encoding and decoding processes. Based on the guided mask, we introduce a new mask conditional stereo entropy (MCSE) model, to fully explore the correlation between the stereo images in entropy coding. In the decoder, we develop a stereo decoding module to simultaneously decode the stereo images and enhance their compression quality. Experimental results show that our MASIC significantly advances the performance of SIC both quantitatively and qualitatively on a variety of datasets, and is robust to the change of parallax level between stereo images. The software codes are available at https://github.com/eecoder-dyf/MASIC.

PyTorch Deep Learning for Food Image Classification with Food Dataset

Classification of food images is crucial in today's increasingly digitally connected world. In the rapidly evolving mobile applications and social media era, the demand for an automated system that can recognize food types from an image is intensifying. This study employs deep learning and the PyTorch framework to develop a dependable and efficient solution for classifying food images. This research is motivated by the growing complexity of food introduction challenges. The primary challenge is improving the accuracy of food type recognition and overcoming variations in the visual presentation of food, such as lighting, shooting angles, and proportional and textural differences. Convolutional Neural Networks (CNN) are effective for image classification and are incorporated into the methods utilized. In addition, we employ ResNet101 transfer learning techniques to capitalize on the knowledge of trained models for large image datasets. The primary objective of this study is to develop a food image classification model that is accurate, training-efficient, and capable of accurately recognizing various types of food. In testing and evaluation, the developed model could realize multiple types of food with satisfactory accuracy. The accuracy of training reached 99.35%, while the accuracy of testing reached 94.65%. This study also reveals how Resnet101 transfer learning is utilized by deep learning technology.

Synplex: In silico modelling of the tumor microenvironment from multiplex images.

Multiplex immunofluorescence is a novel, high-content imaging technique that allows simultaneous in situ labeling of multiple tissue antigens. This technique is of growing relevance in the study of the tumor microenvironment, and the discovery of biomarkers of disease progression or response to immune-based therapies. Given the number of markers and the potential complexity of the spatial interactions involved, the analysis of these images requires the use of machine learning tools that rely for their training on the availability of large image datasets, extremely laborious to annotate. We present Synplex, a computer simulator of multiplexed immunofluorescence images from user-defined parameters: i. cell phenotypes, defined by the level of expression of markers and morphological parameters; ii. cellular neighborhoods based on the spatial association of cell phenotypes; and iii. interactions between cellular neighborhoods. We validate Synplex by generating synthetic tissues that accurately simulate real cancer cohorts with underlying differences in the composition of their tumor microenvironment and show proof-of-principle examples of how Synplex could be used for data augmentation when training machine learning models, and for the in silico selection of clinically relevant biomarkers. Synplex is publicly available at https://github.com/djimenezsanchez/Synplex.

A Brainwide Risk Score for Psychiatric Disorder Evaluated in a Large Adolescent Population Reveals Increased Divergence Among Higher-Risk Groups Relative to Control Participants

BackgroundAccurate psychiatric risk assessment requires biomarkers that are both stable and adaptable to development. Functional network connectivity (FNC), which steadily reconfigures over time, potentially contains abundant information to assess psychiatric risks. However, the absence of suitable analytical methodologies has constrained this area of investigation. MethodsWe investigated the brainwide risk score (BRS), a novel FNC-based metric that contrasts the relative distances of an individual’s FNC to that of psychiatric disorders versus healthy control references. To generate group-level disorder and healthy control references, we utilized a large brain imaging dataset containing 5231 total individuals diagnosed with schizophrenia, autism spectrum disorder, major depressive disorder, and bipolar disorder and their corresponding healthy control individuals. The BRS metric was employed to assess the psychiatric risk in 2 new datasets: Adolescent Brain Cognitive Development (ABCD) Study (n = 8191) and Human Connectome Project Early Psychosis (n = 170). ResultsThe BRS revealed a clear, reproducible gradient of FNC patterns from low to high risk for each psychiatric disorder in unaffected adolescents. We found that low-risk ABCD Study adolescent FNC patterns for each disorder were strongly present in over 25% of the ABCD Study participants and homogeneous, whereas high-risk patterns of each psychiatric disorder were strongly present in about 1% of ABCD Study participants and heterogeneous. The BRS also showed its effectiveness in predicting psychosis scores and distinguishing individuals with early psychosis from healthy control individuals. ConclusionsThe BRS could be a new image-based tool for assessing psychiatric vulnerability over time and in unaffected individuals, and it could also serve as a potential biomarker, facilitating early screening and monitoring interventions.

Medimatrix: innovative pre-training of grayscale images for rheumatoid arthritis diagnosis revolutionises medical image classification.

Efficient and accurate medical image classification (MIC) methods face two major challenges: (1) high similarity between images of different disease classes; and (2) generating large medical image datasets for training deep neural networks is challenging due to privacy restrictions and the need for expert ground truth annotations. In this paper, we introduce a novel deep learning method called pre-training grayscale images with supervised learning for MIC (MediMatrix). Instead of pre-training on color ImageNet, our approach uses MediMatrix on grayscale ImageNet. To improve the performance of the network, we introduce ShuffleAttention (SA), a self-attention mechanism. By combining SA with the multiple residual structure (ResSA block) and replacing short-cut connections with dense residual connections between corresponding layers (densepath), our network can dynamically adjust channel attention weights and receive image inputs of different sizes, resulting in improved feature representation and better discrimination of similarities between different categories. MediMatrix effectively classifies X-ray images of rheumatoid arthritis (RA), enabling efficient screening without the need for expert analysis or invasive testing. Through extensive experiments, we demonstrate the superiority of MediMatrix over state-of-the-art methods and that color is not critical for rich natural image classification. Our results highlight the potential of computer-aided diagnosis combined with MediMatrix as a valuable screening tool for early detection and intervention in RA.

Deep learning enhanced achromatic imaging with a singlet flat lens.

Correction of chromatic aberration is an important issue in color imaging and display. However, realizing broadband achromatic imaging by a singlet lens with high comprehensive performance still remains challenging, though many achromatic flat lenses have been reported recently. Here, we propose a deep-learning-enhanced singlet planar imaging system, implemented by a 3 mm-diameter achromatic flat lens, to achieve relatively high-quality achromatic imaging in the visible. By utilizing a multi-scale convolutional neural network (CNN) imposed to an achromatic multi-level diffractive lens (AMDL), the white light imaging qualities are significantly improved in both indoor and outdoor scenarios. Our experiments are fulfilled via a large paired imaging dataset with respect to a 3 mm-diameter AMDL, which guaranteed with achromatism in a broad wavelength range (400-1100 nm) but a relative low efficiency (∼45%). After our CNN enhancement, the imaging qualities are improved by ∼2 dB, showing competitive achromatic and high-quality imaging with a singlet lens for practical applications.

An introduction to artificial intelligence in machine vision for postharvest detection of disorders in horticultural products

Horticultural products may suffer from surface, subsurface, and internal disorders. These lower the market value and may lead to consumer dissatisfaction. Today’s postharvest industry has a high interest in implementing technologies that allow efficient sorting of disordered products to avoid food loss, and to improve market value and consumer appreciation. Inspection by machine vision using RGB imaging and more advanced techniques such as spectral cameras, X-ray and magnetic resonance imaging (MRI), is increasingly implemented to non-destructively analyze disorders in horticultural products. However, for objective, robust and reliable detection of symptoms of disorders that are often even not obvious to detect by human vision, powerful data analysis methods are essential. In recent years, image analysis by artificial intelligence (AI) has evolved from traditional machine learning (ML) to advanced deep learning (DL), which is reaching impressive detection abilities. The aim of this article is to introduce recent developments in AI for extracting information on postharvest disorders from complex image data obtained by modern imaging systems to the non-AI expert. The main progress in AI for postharvest applications is the increasing adoption of available convolutional neural networks (CNNs), usually developed and tested with large open source image datasets elsewhere, to postharvest image data. Challenges in the design of DL models are identified, such as the need for large quantities of (labeled) data, the model explainability, and generalizability. Although proposed innovations often result in excellent detection abilities, their usability in an industrial context should be further considered, where uncontrollable parameters may negatively impact the performance.