HAB-Net: Bidirectional CNN-Transformer With Hierarchical Alignment for Remote Sensing Scene Classification

This paper proposes a combined framework of CNN+RFC to brain tumor categorization/classification using MRI (Magnetic-Resonance Imaging) images, which combines both CNN (Convolution Neural Networks) and RFC (Random Forest Classification). Preprocessing, Feature bring-out, and Categorization are the three phases of the proposed framework. In the first step, we use the Gaussian Filter Method on the data set then we combine the original data set with processed data in parallel. The feature extraction of magnetic resonance imaging was performed automatically by CNN in the second step. We also called such a type of process in this paper as non-hand-crafted feature extraction. Several classification algorithms, including RFC (Random Forest Classifier), KNN (K-Nearest Neighbor Classifier), DT (Decision Tree Classifier), SVM (Support Vector Machine Classifier), and NB (Naïve Bayes Classifier), are used in the final step. The extracted features from the CNN model are then given to the classifier algorithms, which predict Glioma tumor, Pituitary tumor, Meningioma tumor, and no tumor as a result of the testing data set. Experiments are carried out on an open data set of images selected for classification from the Kaggle databases. This data set is very complex since it contains images of brain tumor with different angles and different depths. We don't alter this data set at all. We make a separate CSV file that contains testing images' name and their specification. Using this proposed approach, we were able to achieve 99.61% accuracy on the training data set, 92.16% on the validation data, and 71.2% on the CSV/testing data.

In an era where COVID-19's varied radiological presentations challenge diagnostic precision, this paper introduces a novel fusion of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models to refine detection capabilities. The paper capitalizes on an expansive radiography database, utilizing lung segmentation and data augmentation to bolster the training set. Advanced CNN models extract salient features, which are then classified through meticulously optimized machine learning algorithms. The synergy between diverse CNN architectures and Bayesian Optimization markedly enhances hyperparameter tuning, culminating in a significant uplift in diagnostic accuracy. The integration of LSTM models further allows for nuanced severity assessments, paving the way for more personalized and effective pandemic management. This research heralds a new frontier in medical diagnostics, promising a leap in the accuracy of pandemic detection.

In this paper, a novel multimodal convolutional neural network (CNN) fusion architecture is proposed for heart sound signal classification. Instead of using features from just one domain, general frequency features as well as Mel domain features are extracted from the raw heart sound. The multimodal CNN fusion architecture is individually trained based on the feature maps resulting from various feature extraction methods. These feature maps are then merged for optimizing the diversified extracted features. The proposed method provides an opportunity to explore the optimal selection of features for heart sound classification. Extensive experimentations are carried out, showing that an outstanding accuracy of 98.5% is achieved by the multimodal CNN architecture, which outperforms the other state-of-the-art approaches.

This study aim was to explore the application effect of computed tomography (CT) image segmentation based on deep learning algorithm in the diagnosis of lung cancer. In this study, a two-dimensional (2D) convolutional neural network (CNN) and three-dimensional (3D) CNN fusion model was constructed firstly. Subsequently, 60 patients with lung cancer were randomly divided into a control group and an intervention group to receive perioperative routine nursing and rehabilitation nursing, respectively. The results revealed that the Dice value (0.876), sensitivity (0.849), and positive predictive value (PPV) (0.875) of the hybrid feature fusion model (HFFM) constructed in this study for lung cancer CT image segmentation were higher than those of other models, and the accuracy rate for lung cancer diagnosis was 96.7%. After nursing intervention, the partial arterial oxygen pressure (PaO2) and partial arterial carbon dioxide pressure (PaCO2) in the control group were 80.54 mmHg and 39.81 mmHg, respectively, while those in the intervention group were 83.09 mmHg and 36.75 mmHg, respectively. After intervention, the maximal voluntary ventilation (MVV) %, forced vital capacity (FVC) %, and forced expiratory volume in 1 sec (FEV1) %) in the intervention group were 76.03%, 82.14%, and 89.76%, respectively. It suggested that compared with the control group, the pulmonary function indexes of the intervention group improved significantly after nursing intervention ( P < 0.05). In summary, the HFFM constructed in this study can be used for segmentation and classification of CT images of lung cancer patients, which can improve the accuracy of diagnosis and help improve the lung function and quality of life of patients.

As cyberattacks become more advanced, conventional centralized threat intelligence models often fail to keep up with these threats’ growing complexity and frequency, highlighting the requirement for innovative approaches to strengthen cybersecurity resilience. Federated learning (FL), a decentralized machine learning (ML) model, provides a promising solution by permitting spread objects to train techniques on local data collaboratively without distributing sensitive data. The efficiency of FL in enhancing attack intelligence skills emphasizes its probability of driving a novel period of robust and privacy-protecting cybersecurity practices. Furthermore, combining FL into cybersecurity structures can strengthen attack intelligence models by permitting real upgrades and adaptive learning mechanisms. Recently, ML and Deep Learning (DL) approaches have drawn the study community to advance security solutions for cyberattack defence mechanism models. Conventional ML and DL techniques that function with data kept on a federal server increase the main privacy issues for user information. This manuscript presents a Cyberattack Defence Mechanism System for Federated Learning Framework using Attention Induced Deep Convolution Neural Networks (CDMFL-AIDCNN) technique. The CDMFL-AIDCNN model presents an improved structure incorporating self-guided FL with attack intelligence to improve defence mechanisms across varied cybersecurity applications in distributed systems. Initially, the data preprocessing stage utilizes Z-score normalization to transform input data into a beneficial format. The Dung Beetle Optimization (DBO) technique is used in the feature selection process to identify the most relevant and non-redundant features. Furthermore, the fusion of convolutional neural networks, bidirectional long short-term memory, gated recurrent units, and attention (CBLG-A) models are employed to classify cyberattack defence mechanisms. Finally, the parameter tuning of the CBLG-A approach is performed by the growth optimizer (GO) approach. The CDMFL-AIDCNN technique is extensively analyzed using the CIC-IDS-2017 and UNSW-NB15 datasets. The comparison analysis of the CDMFL-AIDCNN technique portrayed a superior accuracy value of 99.07% and 98.64% under the CIC-IDS-2017 and UNSW-NB15 datasets.

This paper offers a new methodology for the detection of casting defects through the fusion of Convolutional Neural Networks (CNNs) and Auto-Encoders to extract features and detect anomalies. Through Utilization of a heterogeneous dataset, the model outperforms conventional methods of defect detection, with a 91% accuracy. Contrary to visual inspection and simple algorithms, this deep learning technique driven by data enhances detection accuracy and minimizes false positives. The model is versatile in various casting situations and can be used as a scalable quality control tool. Future work will involve improving model performance, solving data imbalance issues, and applying it to more general manufacturing processes.

Humans have an excellent ability to recognize freehand sketch drawings despite their abstract and sparse structures. Understanding freehand sketches with automated methods is a challenging task due to the diversity and abstract structures of these sketches. In this paper, we propose an efficient freehand sketch recognition scheme, which is based on the feature-level fusion of Convolutional Neural Networks (CNNs) in the transfer learning context. Specifically, we analyse different layer performances of distinct ImageNet pretrained CNNs and combine best performing layer features within the CNN-SVM pipeline for recognition. We also employ Principal Component Analysis (PCA) to reduce the fused deep feature dimensions to ensure the efficiency of the recognition application on the limited-capacity devices. We perform evaluations on two real sketch benchmark datasets, namely the Sketchy and the TU-Berlin to show the effectiveness of the proposed scheme. Our experimental results show that, the feature-level fusion scheme with the PCA achieves a recognition accuracy of 97.91% and 72.5% on the Sketchy and TU-Berlin datasets, respectively. This result is promising when compared with the human recognition accuracy of 73.1% on the TU-Berlin dataset. We also develop a sketch recognition application for smart devices to demonstrate the proposed scheme.

This study presents a novel approach using Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) to generate synthetic electroencephalography (EEG) and electrocardiogram (ECG) waveforms. The synthetic EEG data represent concentration and relaxation mental states, while the synthetic ECG data correspond to normal and abnormal states. By addressing the challenges of limited biophysical data, including privacy concerns and restricted volunteer availability, our model generates realistic synthetic waveforms learned from real data. Combining real and synthetic datasets improved classification accuracy from 92% to 98.45%, highlighting the benefits of dataset augmentation for machine learning performance. The WGAN-GP model achieved 96.84% classification accuracy for synthetic EEG data representing relaxation states and optimal accuracy for concentration states when classified using a fusion of convolutional neural networks (CNNs). A 50% combination of synthetic and real EEG data yielded the highest accuracy of 98.48%. For EEG signals, the real dataset consisted of 60-s recordings across four channels (TP9, AF7, AF8, and TP10) from four individuals, providing approximately 15,000 data points per subject per state. For ECG signals, the dataset contained 1200 real samples, each comprising 140 data points, representing normal and abnormal states. WGAN-GP outperformed a basic generative adversarial network (GAN) in generating reliable synthetic data. For ECG data, a support vector machine (SVM) classifier achieved an accuracy of 98% with real data and 95.8% with synthetic data. Synthetic ECG data improved the random forest (RF) classifier’s accuracy from 97% with real data alone to 98.40% when combined with synthetic data. Statistical significance was assessed using the Wilcoxon signed-rank test, demonstrating the robustness of the WGAN-GP model. Techniques such as discrete wavelet transform, downsampling, and upsampling were employed to enhance data quality. This method shows significant potential in addressing biophysical data scarcity and advancing applications in assistive technologies, human-robot interaction, and mental health monitoring, among other medical applications.

Remote sensing scene classification converts remote sensing images into classification information to support high-level applications, so it is a fundamental problem in the field of remote sensing. In recent years, many convolutional neural network (CNN)-based methods have achieved impressive results in remote sensing scene classification, but they have two problems in extracting remote sensing scene features: (1) fixed-shape convolutional kernels cannot effectively extract features from remote sensing scenes with complex shapes and diverse distributions; (2) the features extracted by CNN contain a large number of redundant and invalid information. To solve these problems, this paper constructs a deformable convolutional neural network to adapt the convolutional sampling positions to the shape of objects in the remote sensing scene. Meanwhile, the spatial and channel attention mechanisms are used to focus on the effective features while suppressing the invalid ones. The experimental results indicate that the proposed method is competitive to the state-of-the-art methods on three remote sensing scene classification datasets (UCM, NWPU, and AID).

To overcome the shortcomings of traditional manual counting methods, which are labor-intensive, resource-consuming, and inefficient, this study introduces a computer-based counting model. This model integrates convolutional neural networks (CNNs) with Transformer networks to efficiently recognize and count specific target objects in large-scale data scenarios. This approach leverages CNNs for local feature extraction and incorporates Transformer networks to capture long-range global information, achieving a synergistic effect. The methodology includes key steps such as “CNN for feature extraction and Transformer for global attention.” The experiment outcomes show that the model has an average absolute error of 10.13, a root mean square error of 12.08, an average counting accuracy of 98.6%, a peak signal-to-noise ratio of 23.75, a structural similarity of 0.933, a coefficient of determination of 0.901, an average counting time of about 6.58ms per image, and a parameter count of 3.21 in target counting. It can also recognize and respond well to high complexity scenes while maintaining high accuracy. Compared to the CNN model, the research model reduces the error rate by 13.4%, indicating that the fusion of CNN and Transformer networks is effective in object counting for computer vision tasks. This result indicates that the model integrating convolutional neural networks and fully self attention networks can be well applied to computer recognition and object counting.

The precipitation oxygen isotopic composition is a useful environmental tracer for climatic and hydrological studies. However, the observed precipitation oxygen is limited at both temporal and spatial scales. Isotope-equipped general circulation models (iGCMs) can compensate for the temporal and spatial discontinuity of observation networks, but they suffer from coarse spatial resolutions and systematic biases. The objective of this study is to build a high-resolution precipitation oxygen isoscape in China for a period of 148 years by integrating observed and iGCMs-simulated precipitation oxygen isotope composition (δ18Op) using data fusion and bias correction methods. The temporal and spatial resolutions are month and 50–60 km for the isoscape, respectively. Prior to building the oxygen isoscape, the performance of two bias correction methods (BCMs) and three data fusion methods (DFMs) is compared after post-processing of eight iGCM simulations. Results show that the outputs of the Convolutional Neural Networks (CNN) fusion method exhibit the strongest correlation with observations with correlation coefficient mostly larger than 0.8, and the smallest bias with root mean square error mostly smaller than 2 ‰. The other two DFMs also perform slightly better than the two BCMs, which show similar performance. Thus, precipitation oxygen isoscape is generated by using the CNN fusion method for the 1969–2007 period in which all iGCMs have output and by using the bias correction methods for the remaining years. Based on the precipitation oxygen isoscape, the spatiotemporal patterns of δ18Op across China are investigated. The generated isoscape shows similar spatial and temporal distribution characteristics to observations. In general, the distribution pattern of δ18Op is consistent with the temperature effect in northern China, and with the precipitation amount effect in southern China, and be more specific in the Qinghai-Tibet Plateau of China. The δ18Op time series mirrors a fluctuating upward trend of the temperature or precipitation in most regions of China. The temporal and spatial distribution characteristics of the generated isoscape are consistent with the characteristics of atmospheric circulation and climate change, indicating successful assimilation and extension of the observed precipitation oxygen isotopes in time and space. Overall, the built isoscape is reliable and useful for providing strong support for tracking atmospheric and hydrological processes. The dataset is available in Zenodo at https://doi.org/10.5281/zenodo.5703811 (Chen et al., 2021).

The precipitation oxygen isotopic composition is a useful environmental tracer for climatic and hydrological studies. However, the observed precipitation oxygen is limited at both temporal and spatial scales. Isotope-equipped general circulation models (iGCMs) can compensate for the temporal and spatial discontinuity of observation networks, but they suffer from coarse spatial resolutions and systematic biases. The objective of this study is to build a high-resolution precipitation oxygen isoscape in China for a period of 148 years by integrating observed and iGCMs-simulated precipitation oxygen isotope composition (δ18Op) using data fusion and bias correction methods. The temporal and spatial resolutions are month and 50–60 km for the isoscape, respectively. Prior to building the oxygen isoscape, the performance of two bias correction methods (BCMs) and three data fusion methods (DFMs) is compared after post-processing of eight iGCM simulations. Results show that the outputs of the Convolutional Neural Networks (CNN) fusion method exhibit the strongest correlation with observations with correlation coefficient mostly larger than 0.8, and the smallest bias with root mean square error mostly smaller than 2 ‰. The other two DFMs also perform slightly better than the two BCMs, which show similar performance. Thus, precipitation oxygen isoscape is generated by using the CNN fusion method for the 1969–2007 period in which all iGCMs have output and by using the bias correction methods for the remaining years. Based on the precipitation oxygen isoscape, the spatiotemporal patterns of δ18Op across China are investigated. The generated isoscape shows similar spatial and temporal distribution characteristics to observations. In general, the distribution pattern of δ18Op is consistent with the temperature effect in northern China, and with the precipitation amount effect in southern China, and be more specific in the Qinghai-Tibet Plateau of China. The δ18Op time series mirrors a fluctuating upward trend of the temperature or precipitation in most regions of China. The temporal and spatial distribution characteristics of the generated isoscape are consistent with the characteristics of atmospheric circulation and climate change, indicating successful assimilation and extension of the observed precipitation oxygen isotopes in time and space. Overall, the built isoscape is reliable and useful for providing strong support for tracking atmospheric and hydrological processes. The dataset is available in Zenodo at https://doi.org/10.5281/zenodo.5703811 (Chen et al., 2021).

Learning powerful discriminative features is the key for remote sensing scene classification. Most existing approaches based on convolutional neural network (CNN) have achieved great results. However, they mainly focus on global-based visual features while ignoring object-based location features, which is important for large-scale scene classification. There are a large number of scene-related ground objects in remote sensing images, as well as Graph convolutional network (GCN) has the potential to capture the dependencies among objects. This article introduces a novel two-stream architecture that combines global-based visual features and object-based location features, so as to improve the feature representation capability. First, we extract appearance visual features from whole scene image based on CNN. Second, we detect ground objects and construct a graph to learn the spatial location features based on GCN. As a result, the network can jointly capture appearance visual information and spatial location information. To the best of authors' knowledge, we are the first to investigate the dependencies among objects in remote sensing scene classification task. Extensive experiments on two datasets show that our framework improves the discriminative ability of features and achieves competitive accuracy against other state-of-the-art approaches.

The pervasive issue of online sexism continues to pose significant challenges, fostering environments characterized by toxicity and perpetuating harmful societal norms. In response, this paper presents an approach for the discovery of sexist statements employing convolutional neural networks (CNNs), Word Embeddings, and data augmentation techniques. Through the fusion of CNNs’ capacity for hierarchical feature extraction with the semantic representations afforded by Word Embeddings, our method achieves exemplary discrimination performance. Additionally, the incorporation of data augmentation enriches the training dataset, thereby augmenting model generalization and resilience. Empirical evaluation on a larger dataset of statements demonstrates the efficacy of our approach, surpassing many baseline approaches in terms of discovery accuracy, precision, recall and F1-score.

Convolutional Neural Networks have been widely used in remote sensing scene classification. Since this kind of model needs a large number of training samples containing data category information, a Generative Adversarial Network (GAN) is usually used to address the problem of lack of samples. However, GAN mainly generates scene data samples that do not contain category information. To address this problem, a novel supervised adversarial Lie Group feature learning network is proposed. In the case of limited data samples, the model can effectively generate data samples with category information. There are two main differences between our method and the traditional GAN. First, our model takes category information and data samples as the input of the model and optimizes the constraint of category information in the loss function, so that data samples containing category information can be generated. Secondly, the object scale sample generation strategy is introduced, which can generate data samples of different scales and ensure that the generated data samples contain richer feature information. After large-scale experiments on two publicly available and challenging datasets, it is found that our method can achieve better scene classification accuracy even with limited data samples.