Search

Visual Question Answering (VQA) is a promising technology that has the potential to revolutionize the medical field by enabling computers to respond to questions about medical images. VQA holds great potential to transform medical imaging, but several obstacles stand in the way, effective medical VQA model development is hampered by problems like the scarcity of easily accessible medical datasets, complicated medical scenarios, and the complexity of the medical images. To contribute to the advancement of VQA models in the medical field, our research undertakes two key initiatives. Firstly, we introduce a novel dataset for medical VQA, derived from an existing dataset of blood cell images, Secondly, our study proposes a VQA model that is specifically designed to classify images of microscopic blood cells. We use pre-trained transformers like Electra, BERT, and DistilBERT to extract textual features in combination with Vision Transformer (ViT) to extract visual features from images, and then combine textual and visual features and then apply classifiers like Linear SVM and Quadratic SVM. Experimental results show that The Electra & ViT model surpasses other models by achieving high scores across multiple evaluation metrics, including a WUPS score of 90.09%, accuracy of 89.63%, F1-Score of 64.03%, precision of 63.42%, and recall of 65.23%.

Traffic signs have great importance regarding smooth traffic flow and safe driving. However, due to many distractions and capricious factors, spotting and perceiving them may become hazardous. Traffic sign detection and recognition have gained popularity to put an end or to lessen the issue, and massive efforts have been realized in this regard. Despite considerable endeavors put together for traffic sign detection and recognition, there is a lack of attention in this area where these traffic signs contain text in them. A handful of studies may be found in state-of-the-art (SOTA) methods for text-based traffic sign detection, and particularly lesser for text recognition of detected text. The proposed method focuses on developing a robust semi-pipeline intelligent system to detect and understand text from traffic road signs boards in various weather conditions. For this purpose, a customized YOLOv5s is deployed for initial panel detection. Subsequently, MSER with preprocessing techniques is used for localization of text. Finally, OCR with NLP is utilized to recognize the text. The proposed method employed the ASAYAR dataset for training and different datasets for testing. The proposed approach produced satisfactory outcomes on them in contrast with SOTA approaches.

Water, especially drinking water, should be clean and free of disease-causing bacteria because of its critical role in life. However, it isn’t easy to identify and classify them rapidly at an early stage. Primarily, the examination of water is performed manually to check the contamination level. Some researchers have proposed techniques to detect and classify bacteria images, but this field still needs more attention. In this research work, a robust Quantum Convolutional Neural Network (QCNN) classification model is proposed to classify the six major categories of pathogenic bacteria. For the acquisition of pathogen images, different slides are created through the gram-staining process, and then images are captured from those slides. DIBaS is the publicly available dataset that provides these slides captured through gram-staining, which is used to evaluate the proposed methodology. So, in the first step, database preprocessing, small patches are extracted from slide images. However, the extracted patches were not clear and very useful, so the Enhanced Super-Resolution Generative Adversarial Network Model (ESRGAN) was applied to images to improve the image quality of extracted patches. The third step is to extract the deep features and classify bacterial images using the QCNN model, in which the Quantum Convolutional layer is added, and classical data is converted into quantum data to perform classification. Based on the results of classification experiments using the QCNN model, the accuracy is 96.54%.

Skin cancer is a serious disease that affects people all over the world. Melanoma is an aggressive form of skin cancer, and early detection can significantly reduce human mortality. In the United States, approximately 97,610 new cases of melanoma will be diagnosed in 2023. However, challenges such as lesion irregularities, low-contrast lesions, intraclass color similarity, redundant features, and imbalanced datasets make improved recognition accuracy using computerized techniques extremely difficult. This work presented a new framework for skin lesion recognition using data augmentation, deep learning, and explainable artificial intelligence. In the proposed framework, data augmentation is performed at the initial step to increase the dataset size, and then two pretrained deep learning models are employed. Both models have been fine-tuned and trained using deep transfer learning. Both models (Xception and ShuffleNet) utilize the global average pooling layer for deep feature extraction. The analysis of this step shows that some important information is missing; therefore, we performed the fusion. After the fusion process, the computational time was increased; therefore, we developed an improved Butterfly Optimization Algorithm. Using this algorithm, only the best features are selected and classified using machine learning classifiers. In addition, a GradCAM-based visualization is performed to analyze the important region in the image. Two publicly available datasets—ISIC2018 and HAM10000—have been utilized and obtained improved accuracy of 99.3% and 91.5%, respectively. Comparing the proposed framework accuracy with state-of-the-art methods reveals improved and less computational time.

Cancer is one of the leading significant causes of illness and chronic disease worldwide. Skin cancer, particularly melanoma, is becoming a severe health problem due to its rising prevalence. The considerable death rate linked with melanoma requires early detection to receive immediate and successful treatment. Lesion detection and classification are more challenging due to many forms of artifacts such as hairs, noise, and irregularity of lesion shape, color, irrelevant features, and textures. In this work, we proposed a deep-learning architecture for classifying multiclass skin cancer and melanoma detection. The proposed architecture consists of four core steps: image preprocessing, feature extraction and fusion, feature selection, and classification. A novel contrast enhancement technique is proposed based on the image luminance information. After that, two pre-trained deep models, DarkNet-53 and DensNet-201, are modified in terms of a residual block at the end and trained through transfer learning. In the learning process, the Genetic algorithm is applied to select hyperparameters. The resultant features are fused using a two-step approach named serial-harmonic mean. This step increases the accuracy of the correct classification, but some irrelevant information is also observed. Therefore, an algorithm is developed to select the best features called marine predator optimization (MPA) controlled Reyni Entropy. The selected features are finally classified using machine learning classifiers for the final classification. Two datasets, ISIC2018 and ISIC2019, have been selected for the experimental process. On these datasets, the obtained maximum accuracy of 85.4% and 98.80%, respectively. To prove the effectiveness of the proposed methods, a detailed comparison is conducted with several recent techniques and shows the proposed framework outperforms.

Natural scene text classification is considered to be a challenging task because of diversified set of image contents, presence of degradations including noise, low contrast/resolution and the random appearance of foreground (font, style, sizes and orientations) and background properties. Above all, the high dimension of the input image’s feature space is another major problem in such tasks. This work is aimed to tackle these problems and remove redundant and irrelevant features to improve the generalization properties of the classifier. In other words, the selection of a qualitative and discriminative set of features, aiming to reduce dimensionality that helps to achieve a successful pattern classification. In this work, we use a biologically inspired genetic algorithm because crossover employed in such algorithm significantly improve the quality of multimodal discriminative set of features and hence improve the classification accuracy for diversified natural scene text images. The Support Vector Machine (SVM) algorithm is used for classification and the average F-Score is used as fitness function and target condition. First after preprocessing input images, the whole feature space (population) is built using a multimodal feature representation technique. Second, a feature level fusion approach is used to combine the features. Third, to improve the average F-score of the classifier, we apply a meta-heuristic optimization technique using a GA for feature selection. The proposed algorithm is tested on five publically available datasets and the results are compared with various state-of-the-art methods. The obtained results proved that the proposed algorithm performs well while classifying textual and non-textual region with better accuracy than benchmark state-of-the-art algorithms.

Provocative heart disease is related to ventricular arrhythmias (VA). Ventricular tachyarrhythmia is an irregular and fast heart rhythm that emerges from inappropriate electrical impulses in the ventricles of the heart. Different types of arrhythmias are associated with different patterns, which can be identified. An electrocardiogram (ECG) is the major analytical tool used to interpret and record ECG signals. ECG signals are nonlinear and difficult to interpret and analyze. We propose a new deep learning approach for the detection of VA. Initially, the ECG signals are transformed into images that have not been done before. Later, these images are normalized and utilized to train the AlexNet, VGG-16 and Inception-v3 deep learning models. Transfer learning is performed to train a model and extract the deep features from different output layers. After that, the features are fused by a concatenation approach, and the best features are selected using a heuristic entropy calculation approach. Finally, supervised learning classifiers are utilized for final feature classification. The results are evaluated on the MIT-BIH dataset and achieved an accuracy of 97.6% (using Cubic Support Vector Machine as a final stage classifier).

Industry 4.0 smart manufacturing systems are equipped with sensors, smart machines, and intelligent robots. The automated in-plant transportation of manufacturing parts through throwing and catching robots is an attempt to accelerate the transportation process and increase productivity by the optimized utilization of in-plant facilities. Such an approach requires intelligent tracking and prediction of the final 3D catching position of thrown objects, while observing their initial flight trajectory in real-time, by catching robot in order to grasp them accurately. Due to non-deterministic nature of such mechanically thrown objects’ flight, accurate prediction of their complete trajectory is only possible if we accurately observe initial trajectory as well as intelligently predict remaining trajectory. The thrown objects in industry can be of any shape but detecting and accurately predicting interception positions of any shape object is an extremely challenging problem that needs to be solved step by step. In this research work, we only considered spherical shape objects as their3D central position can be easily determined. Our work comprised of development of a 3D simulated environment which enabled us to throw object of any mass, diameter, or surface air friction properties in a controlled internal logistics environment. It also enabled us to throw object with any initial velocity and observe its trajectory by placing a simulated pinhole camera at any place within 3D vicinity of internal logistics. We also employed multi-view geometry among simulated cameras in order to observe trajectories more accurately. Hence, it provided us an ample opportunity of precise experimentation in order to create enormous dataset of thrown object trajectories to train an encoder-decoder bidirectional LSTM deep neural network. The trained neural network has given the best results for accurately predicting trajectory of thrown objects in real time.

Mango fruit is in high demand. So, the timely control of mango plant diseases is necessary to gain high returns. Automated recognition of mango plant leaf diseases is still a challenge as manual disease detection is not a feasible choice in this computerized era due to its high cost and the non-availability of mango experts and the variations in the symptoms. Amongst all the challenges, the segmentation of diseased parts is a big issue, being the pre-requisite for correct recognition and identification. For this purpose, a novel segmentation approach is proposed in this study to segment the diseased part by considering the vein pattern of the leaf. This leaf vein-seg approach segments the vein pattern of the leaf. Afterward, features are extracted and fused using canonical correlation analysis (CCA)-based fusion. As a final identification step, a cubic support vector machine (SVM) is implemented to validate the results. The highest accuracy achieved by this proposed model is 95.5%, which proves that the proposed model is very helpful to mango plant growers for the timely recognition and identification of diseases.

Due to the rapid growth in artificial intelligence (AI) and deep learning (DL) approaches, the security and robustness of the deployed algorithms need to be guaranteed. The security susceptibility of the DL algorithms to adversarial examples has been widely acknowledged. The artificially created examples will lead to different instances negatively identified by the DL models that are humanly considered benign. Practical application in actual physical scenarios with adversarial threats shows their features. Thus, adversarial attacks and defense, including machine learning and its reliability, have drawn growing interest and, in recent years, has been a hot topic of research. We introduce a framework that provides a defensive model against the adversarial speckle-noise attack, the adversarial training, and a feature fusion strategy, which preserves the classification with correct labelling. We evaluate and analyze the adversarial attacks and defenses on the retinal fundus images for the Diabetic Retinopathy recognition problem, which is considered a state-of-the-art endeavor. Results obtained on the retinal fundus images, which are prone to adversarial attacks, are 99% accurate and prove that the proposed defensive model is robust.

Documents are stored in a digital form across several organizations. Printing this amount of data and placing it into folders instead of storing digitally is against the practical, economical, and ecological perspective. An efficient way of retrieving data from digitally stored documents is also required. This article presents a real-time supervised learning technique for document classification based on deep convolutional neural network (DCNN), which aims to reduce the impact of adverse document image issues such as signatures, marks, logo, and handwritten notes. The proposed technique’s major steps include data augmentation, feature extraction using pre-trained neural network models, feature fusion, and feature selection. We propose a novel data augmentation technique, which normalizes the imbalanced dataset using the secondary dataset RVL-CDIP. The DCNN features are extracted using the VGG19 and AlexNet networks. The extracted features are fused, and the fused feature vector is optimized by applying a Pearson correlation coefficient-based technique to select the optimized features while removing the redundant features. The proposed technique is tested on the Tobacco3482 dataset, which gives a classification accuracy of 93.1% using a cubic support vector machine classifier, proving the validity of the proposed technique.

As the number of internet users increases so does the number of malicious attacks using malware. The detection of malicious code is becoming critical, and the existing approaches need to be improved. Here, we propose a feature fusion method to combine the features extracted from pre-trained AlexNet and Inception-v3 deep neural networks with features attained using segmentation-based fractal texture analysis (SFTA) of images representing the malware code. In this work, we use distinctive pre-trained models (AlexNet and Inception-V3) for feature extraction. The purpose of deep convolutional neural network (CNN) feature extraction from two models is to improve the malware classifier accuracy, because both models have characteristics and qualities to extract different features. This technique produces a fusion of features to build a multimodal representation of malicious code that can be used to classify the grayscale images, separating the malware into 25 malware classes. The features that are extracted from malware images are then classified using different variants of support vector machine (SVM), k-nearest neighbor (KNN), decision tree (DT), and other classifiers. To improve the classification results, we also adopted data augmentation based on affine image transforms. The presented method is evaluated on a Malimg malware image dataset, achieving an accuracy of 99.3%, which makes it the best among the competing approaches.