Data Augmentation Research Articles

A Novel Data Augmentation Method for Radiomics Analysis Using Image Perturbations.

Radiomics extracts hundreds of features from medical images to quantitively characterize a region of interest (ROI). When applying radiomics, imbalanced or small dataset issues are commonly addressed using under or over-sampling, the latter being applied directly to the extracted features. Aim of this study is to propose a novel balancing and data augmentation technique by applying perturbations (erosion, dilation, contour randomization) to the ROI in cardiac computed tomography images. From the perturbed ROIs, radiomic features are extracted, thus creating additional samples. This approach was tested addressing the clinical problem of distinguishing cardiac amyloidosis (CA) from aortic stenosis (AS) and hypertrophic cardiomyopathy (HCM). Twenty-one CA, thirty-two AS and twenty-one HCM patients were included in the study. From each original and perturbed ROI, 107 radiomic features were extracted. The CA-AS dataset was balanced using the perturbation-based method along with random over-sampling, adaptive synthetic (ADASYN) and the synthetic minority oversampling technique (SMOTE). The same methods were tested to perform data augmentation dealing with CA and HCM. Features were submitted to robustness, redundancy, and relevance analysis testing five feature selection methods (p-value, least absolute shrinkage and selection operator (LASSO), semi-supervised LASSO, principal component analysis (PCA), semi-supervised PCA). Support vector machine performed the classification tasks, and its performance were evaluated by means of a 10-fold cross-validation. The perturbation-based approach provided the best performances in terms of f1 score and balanced accuracy in both CA-AS (f1 score: 80%, AUC: 0.91) and CA-HCM (f1 score: 86%, AUC: 0.92) classifications. These results suggest that ROI perturbations represent a powerful approach to address both data balancing and augmentation issues.

Data augmentation for generating synthetic electrogastrogram time series.

To address an emerging need for large number of diverse datasets for rigor evaluation of signal processing techniques, we developed and evaluated a new method for generating synthetic electrogastrogram time series. We used electrogastrography (EGG) data from an open database to set model parameters and statistical tests to evaluate synthesized data. Additionally, we illustrated method customization for generating artificial EGG time series alterations caused by the simulator sickness. Proposed data augmentation method generates synthetic EGG data with specified duration, sampling frequency, recording state (postprandial or fasting state), overall noise and breathing artifact injection, and pauses in the gastric rhythm (arrhythmia occurrence) with statistically significant difference between postprandial and fasting states in > 70% cases while not accounting for individual differences. Features obtained from the synthetic EGG signal resembling simulator sickness occurrence displayed expected trends. The code for generation of synthetic EGG time series is not only freely available and can be further customized to assess signal processing algorithms but also may be used to increase data diversity for training artificial intelligence (AI) algorithms. The proposed approach is customized for EGG data synthesis but can be easily utilized for other biosignals with similar nature such as electroencephalogram.

Continuous Glucose, Insulin and Lifestyle Data Augmentation in Artificial Pancreas Using Adaptive Generative and Discriminative Models.

Artificial pancreas requires data from multiple sources for accurate insulin dose estimation. These include data from continuous glucose sensors, past insulin dosage information, meal quantity and time and physical activity data. The effectiveness of closed-loop diabetes management systems might be hampered by the absence of these data caused by device error or lack of compliance by patients. In this study, we demonstrate the effect of output sequence length-driven generative and discriminative model selection in high quality data generation and augmentation. This novel generative adversarial network (GAN) based architecture automatically selects the generator and discriminator architecture based on the desired output sequence length. The proposed model is able to generate glucose, physical activity, meal information data for individual patients. The discriminative scores for Ohio T1DM (2018) dataset were 0.17 ±0.03 (Inputs: CGM, CHO, Insulin) and 0.15 ±0.02 (Inputs: CGM, CHO, Insulin, Heart Rate, Steps) and for Ohio T1D (2020) dataset was 0.16 ±0.02 (Inputs: CGM, CHO, Insulin) and 0.15 ±0.02 (Inputs: CGM, CHO, Insulin, acceleration). A mixture of generated and real data was used to test predictive scores for glucose forecasting models. The best RMSE and MARD achieved for OhioT1DM patients were 17.19 ±3.22 and 7.14 ±1.76 for PH=30min with CGM, CHO, Insulin, heartrate and steps as inputs. Similarly, the RMSE and MARD for real+synthetic data were 15.63 ±2.57 and 5.86 ±1.69 respectively. Compared to existing generative models, we demonstrate that sequence length based architecture selection leads to better synthetic data generation for multiple output sequences (CGM, CHO, Insulin) and forecasting accuracy.

Temporal validity reassessment: commonsense reasoning about information obsoleteness

It is useful for machines to know whether text information remains valid or not for various applications including text comprehension, story understanding, temporal information retrieval, and user state tracking on microblogs as well as via chatbot conversations. This kind of inference is still difficult for current models, including also large language models, as it requires temporal commonsense knowledge and reasoning. We approach in this paper the task of Temporal Validity Reassessment, inspired by traditional natural language reasoning to determine the updates of the temporal validity of text content. The task requires judgment whether actions expressed in a sentence are still ongoing or rather completed, hence, whether the sentence still remains valid or has become obsolete, given the presence of context in the form of a supplementary content such as a follow-up sentence. We first construct our own dataset for this task and train several machine learning models. Then we propose an effective method for learning information from an external knowledge base that gives information regarding temporal commonsense knowledge. Using our prepared dataset, we introduce a machine learning model that incorporates the information from the knowledge base and demonstrate that incorporating external knowledge generally improves the results. We also experiment with different embedding types to represent temporal commonsense knowledge as well as with data augmentation methods to increase the size of our dataset.

Toward robust and high-throughput detection of seed defects in X-ray images via deep learning

BackgroundThe detection of internal defects in seeds via non-destructive imaging techniques is a topic of high interest to optimize the quality of seed lots. In this context, X-ray imaging is especially suited. Recent studies have shown the feasibility of defect detection via deep learning models in 3D tomography images. We demonstrate the possibility of performing such deep learning-based analysis on 2D X-ray radiography for a faster yet robust method via the X-Robustifier pipeline proposed in this article.Results2D X-ray images of both defective and defect-free seeds were acquired. A deep learning model based on state-of-the-art object detection neural networks is proposed. Specific data augmentation techniques are introduced to compensate for the low ratio of defects and increase the robustness to variation of the physical parameters of the X-ray imaging systems. The seed defects were accurately detected (F1-score >90%), surpassing human performance in computation time and error rates. The robustness of these models against the principal distortions commonly found in actual agro-industrial conditions is demonstrated, in particular, the robustness to physical noise, dimensionality reduction and the presence of seed coating.ConclusionThis work provides a full pipeline to automatically detect common defects in seeds via 2D X-ray imaging. The method is illustrated on sugar beet and faba bean and could be efficiently extended to other species via the proposed generic X-ray data processing approach (X-Robustifier). Beyond a simple proof of feasibility, this constitutes important results toward the effective use in the routine of deep learning-based automatic detection of seed defects.

Extensive data engineering to the rescue: building a multi-species katydid detector from unbalanced, atypical training datasets.

Passive acoustic monitoring (PAM) is a powerful tool for studying ecosystems. However, its effective application in tropical environments, particularly for insects, poses distinct challenges. Neotropical katydids produce complex species-specific calls, spanning mere milliseconds to seconds and spread across broad audible and ultrasonic frequencies. However, subtle differences in inter-pulse intervals or central frequencies are often the only discriminatory traits. These extremities, coupled with low source levels and susceptibility to masking by ambient noise, challenge species identification in PAM recordings. This study aimed to develop a deep learning-based solution to automate the recognition of 31 katydid species of interest in a biodiverse Panamanian forest with over 80 katydid species. Besides the innate challenges, our efforts were also encumbered by a limited and imbalanced initial training dataset comprising domain-mismatched recordings. To overcome these, we applied rigorous data engineering, improving input variance through controlled playback re-recordings and by employing physics-based data augmentation techniques, and tuning signal-processing, model and training parameters to produce a custom well-fit solution. Methods developed here are incorporated into Koogu, an open-source Python-based toolbox for developing deep learning-based bioacoustic analysis solutions. The parametric implementations offer a valuable resource, enhancing the capabilities of PAM for studying insects in tropical ecosystems. This article is part of the theme issue 'Towards a toolkit for global insect biodiversity monitoring'.

Chest Xray Medical Diagnosis with Deep Learning

Detecting Disease in Chest X-Rays with Deep Learning Chest X-rays (CXRs) is a common and vital diagnostic tool, but interpreting them accurately requires trained radiologists who can be scarce in many areas. This research explores the potential of deep learning (DL), a type of artificial intelligence (AI), to automatically detect diseases in CXRs. The review explains the basics of DL for CXR analysis, including how deep neural networks (DNNs) work and how transfer learning and data augmentation techniques can improve their performance. It then examines recent studies on applying DNNs to identify common CXR abnormalities like lung nodules, pneumonia, and pneumothorax. This includes exploring multi-class classification, where the model can detect multiple diseases simultaneously. The abstract compares the performance of these techniques with human observers and discusses the challenges of implementing DNN models in clinical practice, particularly their relationship with radiologists. Overall, the research investigates how DL can potentially improve disease detection using CXRs, addressing the need for accurate and accessible diagnostic tools. Key Words: Chest X-rays, Deep Learning, Disease Detection, Artificial Intelligence, Neural Networks

Fabric pilling image segmentation by embedding dual-attention mechanism U-Net network

The initial step in fabric pilling rating is to segment and localize the pilling region, which is a crucial and challenging task. This paper presents a fabric puckering image segmentation method that is integrated into a U-Net network with a dual-attention mechanism. We have enhanced the fully convolutional neural network (U-Net) model by incorporating the dual-attention mechanism. This modification has resulted in a powerful feature extraction capability, enabling the objective and accurate segmentation of the fabric puckering region. We refer to this improved model as the dual-attention U-Net. The network model for fabric pilling feature extraction adopts the improved VGG16 model architecture as its encoding part. The model parameters are initialized with VGG16 pre-training weights to accelerate convergence speed. Second, the feature fusion between the corresponding layers of the encoding part and the decoding part was fed into the dual-attention mechanism module to strengthen the weight values of the fabric pilling region adaptively, which made the model more focused on the fabric pilling target region; Third, the dual-attention U-Net model was trained by data augmentation and migration learning strategies to prevent overfitting; Finally, the performance of the model was evaluated with the collected fabric pilling data set. The results of the experiments indicate that the claimed dual-attention U-Net model is superior to the typical U-Net model in terms of accuracy and precision. The dual-attention U-Net model achieved 99.03% accuracy for fabric pilling segmentation.

Optimizing facial feature extraction and localization using YOLOv5: An empirical analysis of backbone architectures with data augmentation for precise facial region detection

Optimizing facial feature extraction and localization using YOLOv5: An empirical analysis of backbone architectures with data augmentation for precise facial region detection

OFIDA: Object-focused image data augmentation with attention-driven graph convolutional networks.

Image data augmentation plays a crucial role in data augmentation (DA) by increasing the quantity and diversity of labeled training data. However, existing methods have limitations. Notably, techniques like image manipulation, erasing, and mixing can distort images, compromising data quality. Accurate representation of objects without confusion is a challenge in methods like auto augment and feature augmentation. Preserving fine details and spatial relationships also proves difficult in certain techniques, as seen in deep generative models. To address these limitations, we propose OFIDA, an object-focused image data augmentation algorithm. OFIDA implements one-to-many enhancements that not only preserve essential target regions but also elevate the authenticity of simulating real-world settings and data distributions. Specifically, OFIDA utilizes a graph-based structure and object detection to streamline augmentation. Specifically, by leveraging graph properties like connectivity and hierarchy, it captures object essence and context for improved comprehension in real-world scenarios. Then, we introduce DynamicFocusNet, a novel object detection algorithm built on the graph framework. DynamicFocusNet merges dynamic graph convolutions and attention mechanisms to flexibly adjust receptive fields. Finally, the detected target images are extracted to facilitate one-to-many data augmentation. Experimental results validate the superiority of our OFIDA method over state-of-the-art methods across six benchmark datasets.

A deep-learning image registration method for pressure-sensitive paint measurements

Image registration is an essential step in the data processing of pressure-sensitive paint (PSP) measurements. As PSP technology is gradually expanded to increasingly harsh environments, it faces challenges such as severe image misalignment due to model deformations/motions, which pose difficulties to traditional feature-based registration algorithms. To improve registration accuracy and efficiency, we propose an end-to-end image registration method based on deep learning. Initially, a PSP dataset based on PSP images is constructed through data augmentation. Three types of residual network and three training strategies are then adopted to prepare the deep-learning model for automatic image registration. The optimal combination of the residual network and training strategy is selected for validation using fabricated PSP images and outperforms two traditional algorithms (i.e., the Sift and Watershed methods). Finally, the performance of the deep-learning method is compared with that of traditional algorithms adopting a new metric of the overlapping rate for assessing the registration accuracy on experimental PSP images. The results show that the deep learning method outperforms the traditional algorithms in terms of registration accuracy and robustness.

A Hybrid Image Augmentation Technique for User- and Environment-Independent Hand Gesture Recognition Based on Deep Learning

This research stems from the increasing use of hand gestures in various applications, such as sign language recognition to electronic device control. The focus is the importance of accuracy and robustness in recognizing hand gestures to avoid misinterpretation and instruction errors. However, many experiments on hand gesture recognition are conducted in limited laboratory environments, which do not fully reflect the everyday use of hand gestures. Therefore, the importance of an ideal background in hand gesture recognition, involving only the signer without any distracting background, is highlighted. In the real world, the use of hand gestures involves various unique environmental conditions, including differences in background colors, varying lighting conditions, and different hand gesture positions. However, the datasets available to train hand gesture recognition models often lack sufficient variability, thereby hindering the development of accurate and adaptable systems. This research aims to develop a robust hand gesture recognition model capable of operating effectively in diverse real-world environments. By leveraging deep learning-based image augmentation techniques, the study seeks to enhance the accuracy of hand gesture recognition by simulating various environmental conditions. Through data duplication and augmentation methods, including background, geometric, and lighting adjustments, the diversity of the primary dataset is expanded to improve the effectiveness of model training. It is important to note that the utilization of the green screen technique, combined with geometric and lighting augmentation, significantly contributes to the model’s ability to recognize hand gestures accurately. The research results show a significant improvement in accuracy, especially with implementing the proposed green screen technique, underscoring its effectiveness in adapting to various environmental contexts. Additionally, the study emphasizes the importance of adjusting augmentation techniques to the dataset’s characteristics for optimal performance. These findings provide valuable insights into the practical application of hand gesture recognition technology and pave the way for further research in tailoring techniques to datasets with varying complexities and environmental variations.

SCFNet: Lightweight Steel Defect Detection Network Based on Spatial Channel Reorganization and Weighted Jump Fusion

The goal of steel defect detection is to enhance the recognition accuracy and accelerate the detection speed with fewer parameters. However, challenges arise in steel sample detection due to issues such as feature ambiguity, low contrast, and similarity among inter-class features. Moreover, limited computing capability makes it difficult for small and medium-sized enterprises to deploy and utilize networks effectively. Therefore, we propose a novel lightweight steel detection network (SCFNet), which is based on spatial channel reconstruction and deep feature fusion. The network adopts a lightweight and efficient feature extraction module (LEM) for multi-scale feature extraction, enhancing the capability to extract blurry features. Simultaneously, we adopt spatial and channel reconstruction convolution (ScConv) to reconstruct the spatial and channel features of the feature maps, enhancing the spatial localization and semantic representation of defects. Additionally, we adopt the Weighted Bidirectional Feature Pyramid Network (BiFPN) for defect feature fusion, thereby enhancing the capability of the model in detecting low-contrast defects. Finally, we discuss the impact of different data augmentation methods on the model accuracy. Extensive experiments are conducted on the NEU-DET dataset, resulting in a final model achieving an mAP of 81.2%. Remarkably, this model only required 2.01 M parameters and 5.9 GFLOPs of computation. Compared to state-of-the-art object detection algorithms, our approach achieves a higher detection accuracy while requiring fewer computational resources, effectively balancing the model size and detection accuracy.

Contrastive learning with token projection for Omicron pneumonia identification from few-shot chest CT images

IntroductionDeep learning-based methods can promote and save critical time for the diagnosis of pneumonia from computed tomography (CT) images of the chest, where the methods usually rely on large amounts of labeled data to learn good visual representations. However, medical images are difficult to obtain and need to be labeled by professional radiologists.MethodsTo address this issue, a novel contrastive learning model with token projection, namely CoTP, is proposed for improving the diagnostic quality of few-shot chest CT images. Specifically, (1) we utilize solely unlabeled data for fitting CoTP, along with a small number of labeled samples for fine-tuning, (2) we present a new Omicron dataset and modify the data augmentation strategy, i.e., random Poisson noise perturbation for the CT interpretation task, and (3) token projection is utilized to further improve the quality of the global visual representations.ResultsThe ResNet50 pre-trained by CoTP attained accuracy (ACC) of 92.35%, sensitivity (SEN) of 92.96%, precision (PRE) of 91.54%, and the area under the receiver-operating characteristics curve (AUC) of 98.90% on the presented Omicron dataset. On the contrary, the ResNet50 without pre-training achieved ACC, SEN, PRE, and AUC of 77.61, 77.90, 76.69, and 85.66%, respectively.ConclusionExtensive experiments reveal that a model pre-trained by CoTP greatly outperforms that without pre-training. The CoTP can improve the efficacy of diagnosis and reduce the heavy workload of radiologists for screening of Omicron pneumonia.

Classification of Brain Image Tumor using EfficientNet B1-B2 Deep Learning

In this study, a new neural network model (EfficientNet B1-B2) was sought for the detection of brain tumors in magnetic resonance imaging (MRI) images. The primary objective was to achieve high accuracy rates so as to classify the images. The deep learning techniques meticulously processed and increased the data augmentation as much as possible for the EfficientNet B1-B2 models. Our experimental results show an accuracy of 98% in the B1 version in Table II. This provides a potentially optimistic view of the application of artificial intelligence technology to disease diagnosis based on medical image analysis. Nonetheless, we must remind ourselves that the dataset we used has limitations in terms of the challenges it can pose. Although the number of potential variations of actual medical images constitutes a major challenge, it is not the only one. Most medical datasets are unbalanced, contain highly variable noise, have a slow internal structure, and are often small in size. Hence, our end goal is to help stimulate not only the field of brain tumor detection and treatment but also the development of more sophisticated classification models in the health context.

Surface Defect Detection of Aluminum Profiles Based on Multiscale and Self-Attention Mechanisms

To address the various challenges in aluminum surface defect detection, such as multiscale intricacies, sensitivity to lighting variations, occlusion, and noise, this study proposes the AluDef-ClassNet model. Firstly, a Gaussian difference pyramid is utilized to capture multiscale image features. Secondly, a self-attention mechanism is introduced to enhance feature representation. Additionally, an improved residual network structure incorporating dilated convolutions is adopted to increase the receptive field, thereby enhancing the network’s ability to learn from extensive information. A small-scale dataset of high-quality aluminum surface defect images is acquired using a CCD camera. To better tackle the challenges in surface defect detection, advanced deep learning techniques and data augmentation strategies are employed. To address the difficulty of data labeling, a transfer learning approach based on fine-tuning is utilized, leveraging prior knowledge to enhance the efficiency and accuracy of model training. In dataset testing, our model achieved a classification accuracy of 98.01%, demonstrating significant advantages over other classification models.

PnP-GA+: Plug-and-Play Domain Adaptation for Gaze Estimation Using Model Variants.

Appearance-based gaze estimation has garnered increasing attention in recent years. However, deep learning-based gaze estimation models still suffer from suboptimal performance when deployed in new domains, e.g., unseen environments or individuals. In our previous work, we took this challenge for the first time by introducing a plug-and-play method (PnP-GA) to adapt the gaze estimation model to new domains. The core concept of PnP-GA is to leverage the diversity brought by a group of model variants to enhance the adaptability to diverse environments. In this article, we propose the PnP-GA+ by extending our approach to explore the impact of assembling model variants using three additional perspectives: color space, data augmentation, and model structure. Moreover, we propose an intra-group attention module that dynamically optimizes pseudo-labeling during adaptation. Experimental results demonstrate that by directly plugging several existing gaze estimation networks into the PnP-GA+ framework, it outperforms state-of-the-art domain adaptation approaches on four standard gaze domain adaptation tasks on public datasets. Our method consistently enhances cross-domain performance, and its versatility is improved through various ways of assembling the model group.

Speculation and negation identification via unified Machine Reading Comprehension frameworks with lexical and syntactic data augmentation

Speculation and Negation Identification focuses on the extraction of speculative and negative cues and scopes. Previous work relied on complete syntactic trees or simply fed sentences into pre-trained language models, which were confronted with poor generalization within and across datasets, and the limitations of training samples. Accordingly, we build a complete pipeline framework that firstly detects cues and then extracts scopes, and propose unified Machine Reading Comprehension paradigms for both cue detection and scope resolution on several datasets. To tackle the insufficiency of training sets and produce more useful samples with appropriate amount of lexical and syntactic knowledge, we apply data augmentation integrating lexical and syntactic features for scope resolution. Experimental results show that our model achieves higher performance than baselines on several publicly accessible corpora.

Electronic properties prediction enhancement of 36 ternary III-IB-VI alloys using a deep feed-forward neural network

The electronic properties of 36 ternary III-IB-VI monolayers have been investigated using a correlation study, in which a comprehensive study of the band structure of these alloys is provided. To this end, a combination of density functional theory calculations and a deep feed-forward neural network is conducted. Principal component analysis and data augmentation are also implemented as pre-processing. The former reduces the size of features to mitigate redundancy, and the latter can prevent overfitting. The deep feed-forward neural network is then used for the final prediction. During the study, bandgaps of the alloys are predicted indirectly and compared with directly obtained bandgaps. The results of the predictions show a maximum precision of 0.00089 for nMSE using the deep feed-forward neural network, which confirms strong dependencies between the electronic properties of the alloys.

Implementing generative adversarial networks for increasing performance of transmission fault classification

<p>An electrical power system is a network that facilitates the sourcing, transfer, and distribution of electrical energy. In the traditional power system, there are eleven types of faults that can occur in the system. This paper focuses on the classification of these faults over a stretch of 100 kilometres. The dataset used is synthetic and generated from a simulated model using MATLAB/Simulink software. Data augmentation is carried out during training to improve the accuracy of the classification. An indirect training approach through generative adversarial network (GAN) is used to classify these overhead transmission line faults. The random forest (RF) classification is used as the base learning model on the original dataset and it achieves accuracy of 84%. However, the base learner RF when used on GAN model generated augmented faulty data, it performs exceptionally well achieving 99% accuracy. One of the recent state-of-art methods is compared with this approach.</p>