Deep Learning-Based Eye-Writing Recognition with Improved Preprocessing and Data Augmentation Techniques

Most Facial Expression Recognition (FER) systems rely on machine learning approaches that require large databases for an effective training. As these are not easily available, a good solution is to augment the databases with appropriate data augmentation (DA) techniques, which are typically based on either geometric transformation or oversampling augmentations (e.g., generative adversarial networks (GANs)). However, it is not always easy to understand which DA technique may be more convenient for FER systems because most state-of-the-art experiments use different settings which makes the impact of DA techniques not comparable. To advance in this respect, in this paper, we evaluate and compare the impact of using well-established DA techniques on the emotion recognition accuracy of a FER system based on the well-known VGG16 convolutional neural network (CNN). In particular, we consider both geometric transformations and GAN to increase the amount of training images. We performed cross-database evaluations: training with the "augmented" KDEF database and testing with two different databases (CK+ and ExpW). The best results were obtained combining horizontal reflection, translation and GAN, bringing an accuracy increase of approximately 30%. This outperforms alternative approaches, except for the one technique that could however rely on a quite bigger database.

Improving the prediction of extreme wind speed events with generative data augmentation techniques

One of the common challenges in medical artificial intelligence (AI) applications using echocardiography is the lack of image data harmonization. This study aims to improve the prediction accuracy of left ventricular ejection fraction (LVEF) AI models by incorporating data augmentation (DA) techniques to address the variability in image data across different vendor machines. A database comprising 15,770 echocardiographic videos from 3154 patients across five different centers was utilized, with the data acquired using various vendor machines. We prepared datasets specific to GE healthcare (GE), philips (PH), and canon medical systems (CA) vendors, including 1911, 804, and 427 cohorts, respectively. A three-dimensional convolutional neural network (3D-CNN) was trained to predict LVEF, using videos consisting of 20 images per heartbeat as input, with training performed exclusively on data from GE machines (GE-based model). DA techniques, including gamma correction, scaling, median filtering, unsharp masking, translation, rotation, noise correction, and image conversion using a cycle generative adversarial network, were applied. A regression analysis using five different chamber views was performed on the GE test data. The accuracy of LVEF prediction using the GE-based model with DA—specifically gamma correction, scaling, and translation corrections—was excellent. The mean absolute error (root mean square error) in test cohorts was 4.33 (5.58) for GE, 4.42 (5.57) for PH, and 4.89 (6.57) for CA, which were comparable to the results of a model developed using data from all vendors, demonstrating the effectiveness of DA in harmonizing images. This study demonstrates that echocardiographic videos are transferable across vendors and that DA is highly effective in improving LVEF predictions from data acquired using different vendor machines.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-21277-y.

This study introduces a new data augmentation method to improve the corruption robustness of the convolutional neural network (CNN). Many data augmentation methods have been studied to reduce overfitting and to improve the generalization ability of CNNs. One of the most widely used data augmentation techniques is Random Erasing which erases a random rectangle region of an image. Most of the augmentation methods are applied in the spatial domain, but augmentation techniques in the frequency domain are less studied. In this study, the image is processed with 2D discrete Fourier transform (DFT), and then Random Erasing is applied in the frequency domain. Finally, the inversed image generated with inverse DFT is used as an input. As a result, the proposed method leads to the robustness improvement of the model against common corruptions.

Mammography is the primary procedure for breast cancer screening, attempting to reduce breast cancer mortality risk with early detection. Deep learning methods have shown strong applicability to various medical images datasets. Due to paucity of available labeled medical images, accurate computer assisted diagnosis requires intensive data augmentation (DA) techniques, such as geometric/intensity transformations of original images. This data when used along with the training data helps to address the limited medical image dataset collected from various sources. Generative Adversarial Networks (GANs) is one of the DA techniques. GAN trained on images can generate new images that contain many authentic characteristics and look realistic to human observers. Therefore, this paper focuses on overcoming the problem of limited labeled dataset, using Deep Convolution GANs (DCGANs). To analyze the closeness between the original and synthetic images, a visualization tool ImageJ was used. In order to validate the proposed model, a visual Turing test was conducted with the help of medical experts.

In various machine learning tasks, including image classification, data augmentation is widely used to expand the training dataset and prevent overfitting. In traditional data augmentation techniques, images from training data are flipped, distorted, added with some noise, or cropped to create new samples from the original images. This paper presents an improved and more effective data augmentation technique for image classification tasks, called Flip-and-Hide (FnH). Our technique was validated on four well-known datasets, MNIST, FMNIST, CIFAR-10 and CIFAR-100. We found 0.03% and 0.12% improvements on FMNIST and CIFAR-10, respectively, compared to the conventional Hide-and-Seek (HaS) augmentation technique. Moreover, we demonstrated the superiority of the proposed FnH augmentation technique by performing some visualisations. Consequently, our technique is particularly useful for image classification tasks that require limited training data.

The auditory brain circuits are biologically constructed to recognize and localize sounds by encoding a combination of cues that help individuals interpret sounds. The development of computational methods inspired by human capacities has established opportunities for improving machine hearing. Recent studies based on deep learning show that using convolutional recurrent neural networks (CRNNs) is a promising approach for sound event detection and localization in spatial sound. Nevertheless, depending on the sound environment, the performance of these systems is still far from reaching perfect metrics. Therefore, this work intends to boost the performance of state-of-the-art (SOTA) systems by using bio-inspired gammatone auditory filters and intensity vectors (IVs) for the acoustic feature extraction stage, along with the implementation of a temporal convolutional network (TCN) block into a CRNN model, to capture long term dependencies. Three data augmentation techniques are applied to increase the small number of samples in spatial audio datasets. The mentioned stages constitute our proposed Gammatone-based Sound Events Localization and Detection (G-SELD) system, which exceeded the SOTA results on four spatial audio datasets with different levels of acoustical complexity and with up to three sound sources overlapping in time.

Recent deep neural network models trained on smaller and less diverse datasets use data augmentation to alleviate limitations such as overfitting, reduced robustness, and lower generalization. Methods using 3D datasets are among the most common to use data augmentation techniques such as random point drop, scaling, translation, rotations, and jittering. However, these data augmentation techniques are fixed and are often applied to the entire object, ignoring the object’s local geometry. Different local neighborhoods on the object surface hold a different amount of geometric complexity. Applying the same data augmentation techniques at the object level is less effective in augmenting local neighborhoods with complex structures. This paper presents PatchAugment, a data augmentation framework to apply different augmentation techniques to the local neighborhoods. Our experimental studies on PointNet++ and DGCNN models demonstrate the effectiveness of PatchAugment on the task of 3D Point Cloud Classification. We evaluated our technique against these models using four benchmark datasets, ModelNet40 (synthetic), ModelNetlO (synthetic), SHREC’16 (synthetic) and ScanObjectNN (real-world).

The most common causes of blindness in adults worldwide, 2.6% of them are caused by diabetic retinopathy, which is a progressive disease caused by complications of diabetes mellitus. Early detection and prompt treatment help save eyesight. The artificial intelligence technology can provide objective and accurate screening results. Deep learning, especially the Convolutional Neural Network (CNN), which is part of artificial intelligence, has proven successful in solving image problems and is very well used in medical image analysis. CNN works well on large datasets, but it will affect network performance to overfitting in fewer datasets. So to solve the problem of small data, data augmentation techniques can be used. There are various kinds of data augmentation techniques. This study used the CNN method to classify diabetic retinopathy disease, compared several suitable data augmentation techniques for retinal fundus images, and used Contrast Limited Adaptive Histogram Equalization (CLAHE) for image enhancement. This study found that the augmented random zoom technique, together with CLAHE, provided the best accuracy of 98% with 96% sensitivity and 100% specificity.

The agricultural sector is considered as the backbone of almost all economies present in the world. Crop protection is remaining as a crucial factor in agriculture, since it determines the quantity and quality of products and increases the profit of the farmers. In the Indian economy, plantain tree cultivation is playing a major role. Plantain Tree is highly exposed to diseases such as Black Sigatoka/Yellow Sigatoka, Panama, Bunchy top, Moko, chlorosis, etc. This study aims at developing a DL model to facilitate crop protection by accurately identifying the diseases that affect the productivity of plantain trees by using the leaf images. The pre-processing of image data will also help to improve the classification accuracy and training time. The main aim of this study is the research investigation and performance comparison of various data augmentation and feature selection techniques that helps in the classification and detection of diseases in plantain trees. The experiments are carried out in real-time datasets, which are collected from different plantain tree cultivation in different fields of Tamil Nadu. The experimental results reveal the necessity to propose effective preprocessing techniques for improving the accuracy of the classification models.

The leather industry plays a crucial role in the production of high-quality leather products to remain competitive in the market. Due to various reasons such as material handling and turning, leather defects can appear at different stages of the production process. This paper proposed deep learning techniques, specifically transfer learning, deep convolutional neural networks (D-CNN), and ensemble learning for automating leather defect detection and classification. The dataset was taken from the open source Kaggle. The dataset was preprocessed from the beginning. However, we performed Error Level Analysis (ELA) to detect possible changes or fraud in the image. We also used data augmentation techniques (resizing, rescaling, flipping, rotation, zooming, and contrasting) on the dataset to improve the model performance. This study investigated deep learning architectures for automatic leather defect detection, comparing them with seven deep convolutional neural networks (D-CNN) models and transfer learning architectures. Furthermore, we introduce and evaluate two novel ensemble models: MER built on transfer learning architecture (MobileNetV2, EfficientNetB0, and ResNet50) and AZL built on D-CNN architecture (AlexNet, ZfNet, and LeNet). A key contribution of this study is the demonstration of significant performance improvements achieved by strategically integrating squeeze and excitation and label smoothing techniques into transfer learning, D-CNN, and ensemble frameworks. Our proposed MER ensemble method achieves a state-of-the-art of 96.39% for leather defect detection. The proposed model was evaluated using performance metrics including precision, recall, F1-score, AUC, ROC, and Matthews Correlation Coefficient (MCC). Additionally, Grad-CAM was employed for explainable AI to visualize and interpret the model’s decision-making process. Comparative analysis against separate transfer learning and D-CNN models fails to establish the superior performance of our MER ensemble, which highlights the potential to significantly advance automated quality inspection in leather production. This study provides insights into evaluating the best deep learning techniques for leather defect classification, paving the way for more accurate, efficient, and reliable industrial applications. Proposes a robust deep learning-based framework for automated leather defect detection using transfer learning, deep CNNs, and two novel ensemble models (MER and AZL), enhanced with squeeze-and-excitation and label smoothing techniques. Achieves state-of-the-art performance with the proposed MER ensemble model, achieve 96.39% accuracy, and demonstrates superior results across precision, recall, F1-score, AUC–ROC, and MCC. Integrates Error Level Analysis (ELA), extensive data augmentation, and Grad-CAM–based explainable AI to ensure data reliability, improve generalization, and provide transparent visual interpretation of defect regions for industrial quality inspection.

Motor imagery brain-computer interfaces (MI-BCIs) have gained a lot of attention in recent years thanks to their potential to enhance rehabilitation and control of prosthetic devices for individuals with motor disabilities. However, accurate classification of motor imagery signals remains a challenging task due to the high inter-subject variability and non-stationarity in the electroencephalogram (EEG) data. In the context of MI-BCIs, with limited data availability, the acquisition of EEG data can be difficult. In this study, several data augmentation techniques have been compared with the proposed data augmentation technique adaptive cross-subject segment replacement (ACSSR). This technique, in conjunction with the proposed deep learning framework, allows for a combination of similar subject pairs to take advantage of one another and boost the classification performance of MI-BCIs. The proposed framework features a multi-domain feature extractor based on common spatial patterns with a sliding window and a parallel two-branch convolutional neural network. The performance of the proposed methodology has been evaluated on the multi-class BCI Competition IV Dataset 2a through repeated 10-fold cross-validation. Experimental results indicated that the implementation of the ACSSR method (80.47%) in the proposed framework has led to a considerable improvement in the classification performance compared to the classification without data augmentation (77.63%), and other fundamental data augmentation techniques used in the literature. The study contributes to the advancements for the development of effective MI-BCIs by showcasing the ability of the ACSSR method to address the challenges in motor imagery signal classification tasks.

Tuberculosis (TB) is determined as a major health threat resulting in approximately 1.8 million people died in 2015 in most of the low and middle income countries. Many of those deaths could have been prevented if TB had been diagnosed and treated at an earlier stage. Nevertheless, recent advanced diagnosis techniques such as the methods of frontal thoracic radiographs have been still cost prohibitive for mass adoption due to the need of individual analysis of each radiograph by properly experienced radiologists. In addition, current outperformances of deep learning accomplish significant results for classification tasks on diverse domains, but its capability remains limited for tuberculosis detection. Therefore, in this study, we examine the efficiency of deep convolutional neural networks (DCNNs) for detecting TB on chest radiographs using public ChestXray14 as training dataset and Montgomery and Shenzhen as two external testing datasets. Multiple preprocessing techniques, tSNE visualization and data augmentation are first performed. Three different pre-trained DCNNs, namely ResNet152, Inception-ResNet and DenseNet121 models are then used to classify X-ray images as having manifestations of pulmonary TB or as healthy. We observe that appropriate data augmentation techniques are able to further increase accuracies of DCNNs. We achieve the best classifier having an average AUC of 0.95 with DenseNet121 while 0.91 and 0.77 with Inception-ResNet and ResNet121, respectively.

From the early stages of Astronomy, the classification of galaxies has been a conundrum that has left astrophysicists in a situation of quandary. Although, previous methods did a phenomenal job in classifying galaxies but while analysing them, certain inefficiencies had been revealed which cannot be overlooked. The objective had been to conduct an analysis of different types of machine learning techniques that have been used to classify galaxies. This analysis had been conducted on the basis of different attributes taken for different types of classification of galaxies. A method had been proposed to classify galaxies with higher accuracy than previous methods. The configuration for the literature analysis used datasets such as ESO-LV and SDSS and discussed the antecedent techniques for classifying galaxies. It had been inferred that a Convolutional Neural Network with certain data augmentation for irregular Galaxies (Irr) gives the best result of all the algorithms that have been discussed in the literature and its analysis. Owing to the aforementioned, an implementation accentuating the use of deep learning algorithms with certain Data Augmentation techniques and certain different activation functions, named daMCOGCNN (data augmentation-based MOrphological Classifier Galaxy Using Convolutional Neural Networks) had been proposed for morphological classification of galaxies. The dataset comprises of 4614 images from SDSS Image Gallery, Galaxy Zoo challenge and Hubble Image Gallery. The efficient implementation of this method gave a testing accuracy of approximately 98% and 97.92% accuracy had been achieved on a dataset taken from different websites such as AstroBin and other such sources. The model introduced here outperforms its earlier contemporaries.

Deep learning (DL) has been shown to obtain superior results for classification tasks in the field of remote sensing hyperspectral imaging. Superpixel-based techniques can be applied to DL, significantly decreasing training and prediction times, but the results are usually far from satisfactory due to overfitting. Data augmentation techniques alleviate the problem by synthetically generating new samples from an existing dataset in order to improve the generalization capabilities of the classification model. In this paper we propose a novel data augmentation framework in the context of superpixel-based DL called dual-window superpixel (DWS). With DWS, data augmentation is performed over patches centered on the superpixels obtained by the application of simple linear iterative clustering (SLIC) superpixel segmentation. DWS is based on dividing the input patches extracted from the superpixels into two regions and independently applying transformations over them. As a result, four different data augmentation techniques are proposed that can be applied to a superpixel-based CNN classification scheme. An extensive comparison in terms of classification accuracy with other data augmentation techniques from the literature using two datasets is also shown. One of the datasets consists of small hyperspectral small scenes commonly found in the literature. The other consists of large multispectral vegetation scenes of river basins. The experimental results show that the proposed approach increases the overall classification accuracy for the selected datasets. In particular, two of the data augmentation techniques introduced, namely, dual-flip and dual-rotate, obtained the best results.