Deep Neural Network Research Articles

An approach for built-up area extraction using different indices and deep neural network (DNN) model

Mapping urbanization and built-up areas from remotely sensed data poses a formidable challenge, particularly when the spectral reflectance of built-up regions intersects with other land types. To address this, numerous spectral indices have been developed. This paper utilizes multiple indices: Normalized Difference Built-up Index (NDBI), Built-up Area Extraction Index (BAEI), Normalized Built-up Area Index (NBAI), New Built-up Index (NBI), Modified Built-up Index (MBI), Band Ratio for Built-up Area (BRBA), and Normalized Difference Vegetation Index (NDVI) to delineate built-up regions. An intersection approach between BAEI, NBAI, and NDVI refines the methodology, resulting in a final built-up map with 92.5 % accuracy and a 0.848 Kappa coefficient. Subsequently, a Deep Neural Network (DNN) model trained on this map achieves over 95 % accuracy in predicting built-up areas from Landsat 5 imagery, and the resultant built-up map achieved an overall accuracy of 92 % and a Kappa coefficient of 0.85. The proposed methodology demonstrates efficiency for time-series analysis and addresses misclassification in built-up areas. Moreover, the optimization of the DNN model proves effective when meticulous training and validation processes incorporate more precise sample datasets.

A learnable motion preserving pooling for action recognition

Using deep neural networks (DNN) for video understanding tasks is expensive in terms of computation cost. Pooling layers in DNN which are widely used in most vision tasks to resize the spatial dimensions play crucial roles in reducing the computation and memory cost. In video-related tasks, pooling layers are also applied, mostly in the spatial dimension only as the standard average pooling in the temporal domain can significantly reduce its performance. This is because conventional temporal pooling degrades the underlying important motion features in consecutive frames. Such a phenomenon is rarely investigated and most state-of-art methods simply do not adopt temporal pooling, leading to enormous computation costs. In this work, we propose a learnable motion-preserving pooling (MPPool) layer that is able to preserve the general motion progression after the pooling. This pooling layer first locates the frames with the strongest motion features and then keeps these crucial features during pooling. Our experiments demonstrate that MPPool not only reduces the computation cost for video data modeling, but also increases the final prediction accuracy on various motion-centric and appearance-centric datasets.

Securing DNS over HTTPS traffic: a real-time analysis tool

DNS over HTTPS (DoH) is a developing protocol that uses encryption to secure domain name system (DNS) queries within hypertext transfer protocol secure (HTTPS) connections, thereby improving privacy and security while browsing the web. This study involved the development of a live tool that captures and analyzes DoH traffic in order to classify it as either benign or malicious. We employed machine learning (ML) algorithms such as K-Nearest Neighbors (K-NN), random forest (RF), decision tree (DT), deep neural network (DNN), and support vector machine (SVM) to categorize the data. All of the algorithms, namely KNN, RF, and DT, achieved exceptional performance, with F1 scores of 1.0 or above for both precision and recall. The SVM and DNN both achieved exceptionally high scores, with only slight differences in accuracy. This tool employs a voting mechanism to arrive at a definitive classification decision. By integrating with the Mallory tool, it becomes possible to locally resolve DNS, which in turn allows for more accurate simulation of DoH queries. The evaluation results clearly indicate outstanding performance, confirming the tool's effectiveness in analyzing DoH traffic for network security and threat detection purposes.

Real-time Facial State Recognition and Fatigue Analysis based on Deep Neural Networks

Real-time Facial State Recognition and Fatigue Analysis based on Deep Neural Networks

A novel ensemble machine learning exposure model system for ground-level ozone at the national scale: A case of mainland China from 2013 to 2020

In epidemiological research, accurate estimation of historical ground-level ozone (O3) concentrations with enhanced spatiotemporal resolution is crucial for effective exposure assessment. The current state-of-the-art for estimating air pollutant concentrations is a two-stage ensemble method that integrates outputs from multiple machine learning algorithms. Despite its effectiveness, opportunities exist to refine this approach for more precise O3 estimation. In this study, we propose an enhanced ensemble method that incorporates four key strategies. First, we employ high-resolution spatiotemporal predictors derived from prior machine learning studies for refined secondary learning. Second, we use sophisticated algorithms, including categorical gradient boosting, deep neural network, random forest, stochastic variable Gaussian process, transformer, and a combination of convolutional neural network and long short-term memory neural network, as sublearners to enhance learning capabilities. Third, we spatiotemporally split the sample set and then train submodels separately on each subset to eliminate the unobserved spatiotemporal heterogeneity. Finally, we apply a complex machine learning algorithm, rather than the generalized additive model, for integrating sublearner predictions, enabling the capture of intricate nonlinear relationships beyond basic spatiotemporal linear weights. To validate these improvements, we estimated daily maximum 8-h moving average O3 concentrations ([O3]MDA8) across Chinese mainland from 2013 to 2020 at a 1 km spatial resolution. The proposed method demonstrated notable accuracy, achieving an out-of-station determination coefficient (R2) of 0.943 and a root-mean-square error (RMSE) of 10.197 μg/m3. This performance marks a nearly 15% improvement over the best existing Chinese O3 exposure model based on a single algorithm and also surpasses previous studies utilizing traditional ensemble methods for other air pollutants. Our enhanced ensemble approach significantly bolsters the reliability and robustness of future environmental epidemiological studies by further mitigating “misclassification” errors.

Split tensile strength of fiber-reinforced coral aggregate concrete: Deep learning model and experimental validation

Machine learning methods are increasingly becoming an important research field for solving the complex mechanical behavior of concrete materials. A reliable deep learning model was developed based on a dataset consisting of 319 experimental data points to predict the split tensile strength of fiber-reinforced coral aggregate concrete (FRCAC). The feature selection was carried out using correlation analysis, and the neural network structure was optimized by adjusting hyperparameters based on the 10-fold cross-validation (CV). Subsequently, six statistical indicators were employed to evaluate the accuracy and generalization ability of the model. The deep neural network (DNN) model demonstrates excellent performance, with R2, MAE, and MAPE of 0.98, 0.181 MPa, and 0.037 for the training dataset and 0.95, 0.286 MPa, and 0.074 for the test dataset, respectively. Additionally, 24 sets of concrete proportions are designed using the trained DNN model, and the results indicate that the absolute error and relative error between predicted and experimental values are less than 0.4 MPa and 6.8 %, respectively. The feature analysis based on SHapley Additive exPlanations (SHAP) reveals that volume fraction, curing age, water-binder ratio, and the tensile strength of fiber have the highest contribution in estimating the split tensile strength of FRCAC. This research can provide a reference for optimizing the design of FRCAC.

Multiphysics Inverse Design of Frequency-Selective Surface by Data-Physics-Driven Deep Neural Network

Multiphysics Inverse Design of Frequency-Selective Surface by Data-Physics-Driven Deep Neural Network

Predicting antioxidant activity of compounds based on chemical structure using machine learning methods.

Oxidative stress is a well-established risk factor for numerous chronic diseases, emphasizing the need for efficient identification of potent antioxidants. Conventional methods for assessing antioxidant properties are often time-consuming and resource-intensive, typically relying on laborious biochemical assays. In this study, we investigated the applicability of machine learning (ML) algorithms for predicting the antioxidant activity of compounds based solely on their molecular structure. We evaluated the performance of five ML algorithms, Support Vector Machine (SVM), Logistic Regression (LR), XGBoost, Random Forest (RF), and Deep Neural Network (DNN), using a dataset of over 1,900 compounds with experimentally determined antioxidant activity. Both RF and SVM achieved the best overall performance, exhibiting high accuracy (> 0.9) and effectively distinguishing active and inactive compounds with high structural similarity. External validation using natural product data from the BATMAN database confirmed the generalizability of the RF and SVM models. Our results suggest that ML models serve as powerful tools to expedite the discovery of novel antioxidant candidates, potentially streamlining the development of future therapeutic interventions.

Integrated remote sensing and geochemical studies for enhanced prospectivity mapping of porphyry copper deposits: a case study from the Pariz district, Urmia-Dokhtar metallogenic belt, southern Iran

Mapping hydrothermal alteration zones associated with porphyry copper deposits (PCDs) is crucial for identifying new exploration targets on a regional scale. Hydrothermal alteration indicator layers play a fundamental role in recognizing potential areas for PCDs, highlighting the need for precise delineation of these zones and their integration with geochemical and geological data to reduce uncertainty in mapping porphyry copper prospectivity. This study focuses on the Pariz district within the Urmia-Dokhtar Metallogenic Belt (UDMB) in southern Iran, a region known for its significant porphyry copper mineralization. First, logical operator algorithms (LOA) were applied to ASTER remote sensing data to map and distinguish argillic and phyllic alteration zones associated with PCDs. Subsequently, propylitic alteration zones associated with chlorite-epidote and propylitic alteration associated with calcite were also delineated, as were silica-rich hydrothermal alteration zones. Five evidence layers corresponding to these geologic features were generated and weighted with logistic functions, independent of expert judgment and without consideration of the spatial distribution of known mineral occurrences (KMOs). In addition, two layers of information were developed, including multivariate geochemical signatures and proximity to intrusive rocks. The geochemical analysis identified two significant factors associated with porphyry copper mineralization: Factor-I (Zn, Pb, Cu, Sn, B) and Factor-II (Mo, Cu). These factors contributed to a multivariate geochemical signature in addition to the alteration layers derived from remote sensing. Evaluation using prediction- area (P-A) plots and Normalized density index (ND) confirmed the effectiveness of all seven layers for mineral prospectivity mapping (MPM). Geometric average (GA), data-driven index overlay (IO), and deep autoencoder neural network (DEA) integrated these layers, with IO showing superior performance in identifying high potential zones, as indicated by higher prediction rates compared to other methods. Therefore, IO proves to be the most efficient approach for mapping the regional porphyry copper minerals in the Pariz district of the UDMB.

Solutions to elliptic and parabolic problems via finite difference based unsupervised small linear convolutional neural networks

In recent years, there has been a growing interest in leveraging deep learning and neural networks to address scientific problems, particularly in solving partial differential equations (PDEs). However, many neural network-based methods like PINNs rely on auto differentiation and sampling collocation points, leading to a lack of interpretability and lower accuracy than traditional numerical methods. As a result, we propose a fully unsupervised approach, requiring no training data, to estimate finite difference solutions for PDEs directly via small linear convolutional neural networks. Our proposed approach uses substantially fewer parameters than similar finite difference-based approaches while also demonstrating comparable accuracy to the true solution for several selected elliptic and parabolic problems compared to the finite difference method.

Forensic bone age assessment of hand and wrist joint MRI images in Chinese han male adolescents based on deep convolutional neural networks.

In Chinese criminal law, the ages of 12, 14, 16, and 18 years old play a significant role in the determination of criminal responsibility. In this study, we developed an epiphyseal grading system based on magnetic resonance image (MRI) of the hand and wrist for the Chinese Han population and explored the feasibility of employing deep learning techniques for bone age assessment based on MRI of the hand and wrist. This study selected 282 Chinese Han Chinese males aged 6.0-21.0 years old. In the course of our study, we proposed a novel deep learning model for extracting and enhancing MRI hand and wrist bone features to enhance the prediction of target MRI hand and wrist bone age and achieve precise classification of the target MRI and regression of bone age. The evaluation metric for the classification model including precision, specificity, sensitivity, and accuracy, while the evaluation metrics chosen for the regression model are MAE. The epiphyseal grading was used as a supervised method, which effectively solved the problem of unbalanced sample distribution, and the two experts showed strong consistency in the epiphyseal plate grading process. In the classification results, the accuracy in distinguishing between adults and minors was 91.1%, and the lowest accuracy in the three minor classifications (12, 14, and 16 years of age) was 94.6%, 91.1% and 96.4%, respectively. The MAE of the regression results was 1.24 years. In conclusion, the deep learning model proposed enabled the age assessment of hand and wrist bones based on MRI.

Seismic drifts of buildings through deep neural networks circumventing incremental dynamic analysis: Aims and pitfalls for robust and reliable predictions

In the field of seismic response assessment, simplified procedures have long been pursued to circumvent the rigorous, state-of-the-art nonlinear response history analysis (NRHA) and incremental dynamic analysis (IDA) that typically require tedious computations. To this end, this study investigates artificial neural networks (ANN) as prediction models to bypass IDA and quickly and reliably determine the structural drift responses of buildings, without employing surrogate models or analysis techniques that lower the quality of the estimates. Three designs of such prediction models are identified in terms of their scope, implementation and corresponding database, commonly employed in research, while identifying common pitfalls in their design. The investigations involve ten steel-frame multi-story structures, considering cyclic and in-cycle material deterioration and P-Delta phenomena. More than 17-thousand recorded ground motions (GM) are employed to perform IDA on these frame-building models, resulting in a unique database of responses ranging from linear to collapse, with more than 3-million NRHA. Based on this database, a prediction model is developed for each building, capable of predicting the IDA under a GM excitation not included in the database. Additionally, another prediction model is developed that can predict the IDA of a building under a GM excitation, both of which are not included in the database. These investigations show that ANN are excellent tools to circumvent IDA and capture the record-to-record uncertainty by rapidly and reliably predicting structural drifts ranging from linear responses to the collapse limit state, while building-to-building predictions are additionally feasible if the employed database is appropriate.

Deep Variational Network Toward Blind Image Restoration.

Blind image restoration (IR) is a common yet challenging problem in computer vision. Classical model-based methods and recent deep learning (DL)-based methods represent two different methodologies for this problem, each with their own merits and drawbacks. In this paper, we propose a novel blind image restoration method, aiming to integrate both the advantages of them. Specifically, we construct a general Bayesian generative model for the blind IR, which explicitly depicts the degradation process. In this proposed model, a pixel-wise non-i.i.d. Gaussian distribution is employed to fit the image noise. It is with more flexibility than the simple i.i.d. Gaussian or Laplacian distributions as adopted in most of conventional methods, so as to handle more complicated noise types contained in the image degradation. To solve the model, we design a variational inference algorithm where all the expected posteriori distributions are parameterized as deep neural networks to increase their model capability. Notably, such an inference algorithm induces a unified framework to jointly deal with the tasks of degradation estimation and image restoration. Further, the degradation information estimated in the former task is utilized to guide the latter IR process. Experiments on two typical blind IR tasks, namely image denoising and super-resolution, demonstrate that the proposed method achieves superior performance over current state-of-the-arts.

Joint Pruning and Channel-Wise Mixed-Precision Quantization for Efficient Deep Neural Networks

Joint Pruning and Channel-Wise Mixed-Precision Quantization for Efficient Deep Neural Networks

Development of a Machine-learning-based Framework to Assess the Performance of Water Distribution Systems

Recent statistics indicate that the national water supply coverage is approximately 99%, reflecting a high level of service. However, the aging infrastructure continues to cause issues such as pipe failures and water quality concerns, creating operational difficulties. To solve these issues, performance assessments are used to quantitatively evaluate the system conditions and ensure effective maintenance. In Korea, these assessments are based on a scoring system that combines direct and indirect assessments. However, direct assessment of all pipes is limited by cost and time constraints. To address this issue, a deep neural network is used to assess the aging of water distribution systems. This study develops a framework to estimate direct results from indirect data by analyzing the correlation between indirect and direct assessments. Additionally, data augmentation is considered to compensate for the lack of training data in direct assessments, improve the reliability of performance assessments, and support better prioritization in system maintenance.

A statistical journey to DNN, the first trip: From regression to deep neural network

A statistical journey to DNN, the first trip: From regression to deep neural network

DeepWhaleNet: Climate Change-Aware FFT-Based Deep Neural Network for Passive Acoustic Monitoring

Climate change poses severe risks to the survival of many whale populations, whose habitats and migration patterns are affected by environmental changes. To detect these whales effectively, especially in deep-sea environments, we need to use AI-based techniques to handle the acoustic diversity and variability of different species. However, current methods for whale detection are based on pre- and post-processing steps that reduce their efficiency and generalizability. To address this issue, we present DeepWhaleNet, a novel deep-learning model that automates whale detection in Underwater Passive Acoustic Monitoring datasets. DeepWhaleNet simplifies the detection process by extracting relevant features from raw log-power spectrograms and helps protect these threatened species by supporting conservation efforts. Our model uses a larger short-time Fourier transform as input and a custom ResNet-18 architecture for classification, which enables it to separate whale sounds from noise and capture their temporal and spectral characteristics. We evaluate the performance of DeepWhaleNet and show that it surpasses state-of-the-art methods, achieving an 8.3% improvement in the F-1 score and 21% higher average precision of binary relevance than the baseline method. Moreover, our model demonstrates its versatility and suitability for species-specific retrieval problems through an ablation study on multi-label retrieval problems and a 99.1% recall for Blue Whales.

AI-driven predictive maintenance and optimization of renewable energy systems for enhanced operational efficiency and longevity

The rapid growth of renewable energy systems necessitates advanced strategies for maintenance and optimization to ensure long-term operational efficiency and sustainability. Traditional approaches often fall short in predicting failures and optimizing performance across diverse and dynamic renewable energy infrastructures. This study investigates the application of artificial intelligence (AI) techniques for predictive maintenance and optimization of renewable energy systems, with the aim of enhancing operational efficiency and extending system longevity. We employ a combination of machine learning algorithms, including deep neural networks and reinforcement learning, to develop predictive models and optimization strategies. These models are trained on large-scale datasets collected from operational wind farms, solar installations, and hydroelectric plants. Our results demonstrate that AI-driven approaches can predict equipment failures with 92% accuracy, reducing unplanned downtime by 35% compared to traditional methods. Moreover, AI-optimized operational parameters improved overall energy output by 8.5% across the studied systems. The proposed framework also showed adaptability to various environmental conditions and system configurations, suggesting broad applicability across the renewable energy sector. This research underscores the significant potential of AI in revolutionizing maintenance practices and operational strategies in renewable energy systems, paving the way for more reliable, efficient, and sustainable clean energy production.

SSAT: Active Authorization Control and User’s Fingerprint Tracking Framework for DNN IP Protection

As training a high-performance deep neural network (DNN) model requires a large amount of data, powerful computing resources and expert knowledge, protecting well-trained DNN models from intellectual property (IP) infringement has raised serious concerns in recent years. Most existing methods using DNN watermarks to verify the ownership of the models after IP infringement occurs, which is reactive in the sense that they cannot prevent unauthorized users from using the model in the first place. Different from these methods, in this article, we propose an active authorization control and user’s fingerprint tracking method for the IP protection of DNN models by utilizing sample-specific backdoor attack. The proposed method inversely and multiplely exploits sample-specific trigger as the key to implement authorization control for DNN model, in which the generated triggers are imperceptible and sample-specific for clean images. Specifically, a U-Net model is used to generate backdoor instances. Then, the target model is trained on the clean images and backdoor instances, which are inversely labeled as wrong classes and correct classes, respectively. Only authorized users can use the target model normally by pre-processing the clean images through the U-Net model. Moreover, the images processed by the U-Net model will contain unique fingerprint that can be extracted to verify and track the corresponding user’s identity. This article is the first work that utilizes the sample-specific backdoor attack to implement active authorization control and user’s fingerprint management for DNN model under black-box scenarios. Extensive experimental results on ImageNet dataset and YouTube Aligned Face dataset demonstrate that the proposed method is effective in protecting the DNN model from unauthorized usage. Specifically, the protected model has a low inference accuracy (1.00%) for unauthorized users, while maintaining a normal inference accuracy (97.67%) for authorized users. Besides, the proposed method can achieve 100% fingerprint tracking success rates on both the ImageNet and YouTube Aligned Face datasets. Moreover, it is demonstrated that the proposed method is robust against fine-tuning attack, pruning attack, pruning attack with retraining, reverse-engineering attack, adaptive attack, and JPEG compression attack. The code is available at https://github.com/nuaaaisec/SSAT .

Analysis of the Application of Artificial Intelligence in Mahjong

Over the past few years, AI-based models for Mahjong have received a substantial amount of contribution due to the advancements in machine learning and game theory. In this review, these state-of-the-art AI models are considered, including Tjong, Kanachan, Suphx, and Zhejiang University’s developed model. The deployed approaches are disparately represented by reinforcement learning, Monte Carlo tree search (MCTS), and deep neural networks, all of which aim at solving the issues that come with the game’s structure, such as incomplete and overburdening information. The Tjong model complements the power of the transformer architecture with hierarchical decision-making and in turn, brings strategic depth to the gameplay. Kanachan employs Q-learning and MCTS to optimize decision-making, while Suphx combines these methods with the novel ideas of Double Q-learning and Thompson Sampling to achieve higher performance than seen before. The integration of reinforcement learning and a new evaluation model imbues the model with the unique properties of both learning machines and human expertise, namely depth and intelligence. As a result of all these exploits, obstacles still exist, the most notable among them are the management of completely unknown information, handling long-distance games, and the strategic balance between offense and defense. In the future, determinants of probabilistic reasoning, the improvement of deep learning systems, and the prowess of reward shaping will result in AI improvement in Mahjong.