Deep Convolutional Network Research Articles

Drainage Network Generation for Urban Pluvial Flooding (UPF) Using Generative Adversarial Networks (GANs) and GIS Data

Mapping urban pluvial flooding (UPF) in data-scarce regions poses significant challenges, particularly when drainage systems are inadequate or outdated. These limitations hinder effective flood mitigation and risk assessment. This study proposes an innovative approach to address these challenges by integrating deep learning (DL) models with traditional methods. First, deep convolutional generative adversarial networks (DCGANs) were employed to enhance drainage network data generation. Second, deep recurrent neural networks (DRNNs) and multi-criteria decision analysis (MCDA) methods were implemented to assess UPF. The study compared the performance of these approaches, highlighting the potential of DL models in providing more accurate and robust flood mapping outcomes. The methodology was applied to Lahore, Pakistan—a rapidly urbanizing and data-scarce region frequently impacted by UPF during monsoons. High-resolution ALOS PALSAR DEM data were utilized to extract natural drainage networks, while synthetic datasets generated by GANs addressed the lack of historical flood data. Results demonstrated the superiority of DL-based approaches over traditional MCDA methods, showcasing their potential for broader applicability in similar regions worldwide. This research emphasizes the role of DL models in advancing urban flood mapping, providing valuable insights for urban planners and policymakers to mitigate flooding risks and improve resilience in vulnerable regions.

Voltage Prediction of Lithium‐Ion Battery Based on Time Contrastive Learning Model

This study presents a neural network approach to predict the voltage of lithium‐ion batteries. The model employs three voltage data enhancement techniques to improve the feature representation of voltage data. It utilizes a deep dilated convolution network combined with a self‐supervised triplet loss function based on negative sampling. This approach is designed to learn a time series representation method that effectively addresses the issue of missing data. Subsequently, the model leverages a Transformer architecture to extract high‐dimensional characteristics of voltage data. In a battery pack or battery array, monitoring multiple batteries simultaneously allows the encoder part to process the entire sequence in parallel. Finally, voltage prediction is achieved through time contrast learning. The results demonstrate that the model outperforms conventional neural network models across three voltage datasets. Notably, on the NASA dataset, the model reduces the relative error in voltage time series prediction by 2.31% compared to the Transformer model and by 16.27% compared to the long short‐term memory model. In the large‐scale voltage dataset of LiFePO4 battery in an energy storage power station, the performance is even better, and the task of accurate and efficient prediction has been accomplished excellently.

A deep convolutional generative adversarial network (DCGAN) for the fast estimation of pollutant dispersion fields in indoor environments

A deep convolutional generative adversarial network (DCGAN) for the fast estimation of pollutant dispersion fields in indoor environments

Multimodal learning analytics for students behavior prediction using multi-scale dilated deep temporal convolution network with improved chameleon Swarm algorithm

Multimodal learning analytics for students behavior prediction using multi-scale dilated deep temporal convolution network with improved chameleon Swarm algorithm

Enhanced heart disease risk prediction using adaptive botox optimization based deep long-term recurrent convolutional network.

BackgroundHeart disease is the leading cause of death worldwide and predicting it is a complex task requiring extensive expertise. Recent advancements in IoT-based illness prediction have enabled accurate classification using sensor data.ObjectiveThis research introduces a methodology for heart disease classification, integrating advanced data preprocessing, feature selection, and deep learning (DL) techniques tailored for IoT sensor data.MethodsThe work employs Clustering-based Data Imputation and Normalization (CDIN) and Robust Mahalanobis Distance-based Outlier Detection (RMDBOD) for preprocessing, ensuring data quality. Feature selection is achieved using the Improved Binary Quantum-based Avian Navigation Optimization (IBQANO) algorithm, and classification is performed with the Deep Long-Term Recurrent Convolutional Network (DLRCN), fine-tuned using the Adaptive Botox Optimization Algorithm (ABOA).ResultsThe proposed models tested on the Hungarian, UCI, and Cleveland heart disease datasets demonstrate significant improvements over existing methods. Specifically, the Cleveland dataset model achieves an accuracy of 99.72%, while the UCI dataset model achieves an accuracy of 99.41%.ConclusionThis methodology represents a significant advancement in remote healthcare monitoring, crucial for managing conditions like high blood pressure, especially in older adults, offering a reliable and accurate solution for heart disease prediction.

Enhancing Recognition and Categorization of Skin Lesions with Tailored Deep Convolutional Networks and Robust Data Augmentation Techniques

This study introduces deep convolutional neural network-based methods for the detection and classification of skin lesions, enhancing system accuracy through a combination of architectures, pre-processing techniques and data augmentation. Multiple networks, including XceptionNet, DenseNet, MobileNet, NASNet Mobile, and EfficientNet, were evaluated to test deep learning’s potential in complex, multi-class classification tasks. Training these models on pre-processed datasets with optimized hyper-parameters (e.g., batch size, learning rate, and dropout) improved classification precision for early-stage skin cancers. Evaluation measures such as accuracy and loss confirmed high classification efficiency with minimal overfitting, as the validation results aligned closely with training. DenseNet-201 and MobileNet-V3 Large demonstrated strong generalization abilities, whereas EfficientNetV2-B3 and NASNet Mobile achieved the best balance between accuracy and efficiency. The application of different augmentation rates per class also enhanced the handling of imbalanced data, resulting in more accurate large-scale detection. Comprehensive pre-processing ensured balanced class representation, and EfficientNetV2 models achieved exceptional classification accuracy, attributed to their optimized architecture balancing depth, width, and resolution. These models showed high convergence rates and generalization, supporting their suitability for medical imaging tasks using transfer learning.

Image Quality Enhancement using Deep learning-based Convolution Residual Networks Technique

A new approach for cleaning encrypted images is developed using Deep Convolutional Residual Network (Deep ConvResNet) as the proposed method. The aim of this research is to protect encrypted images from noise attacks by utilizing ResNet denoising capabilities. It has been proven that ResNets are successful at cleaning up noise while maintaining the important picture characteristics. This research employs multiple datasets for training and performs a detailed comparative study using peak signal to noise ratio and structure similarity index as well as noise and occlusion and blur attack metrics. Results of the simulation show that the suggested cryptosystem is resistant to the familiar attacks. Filtering based denoising techniques and CNNs are worse than ResNets because ResNets have better efficiency and resilience to occlusion and noise and blur attacks. The graphs of training loss vs. epoch show the convergence pattern of the model during training. Considering its potential use, this methodology is applicable in secure image transmission in different domains such as healthcare and multimedia transmission.

An enhanced CNN with ResNet50 and LSTM deep learning forecasting model for climate change decision making

Climate change poses a significant challenge to wind energy production. It involves long-term, noticeable changes in key climatic factors such as wind power, temperature, wind speed, and wind patterns. Addressing climate change is essential to safeguarding our environment, societies, and economies. In this context, accurately forecasting temperature and wind power becomes crucial for ensuring the stable operation of wind energy systems and for effective power system planning and management. Numerous approaches to wind change forecasting have been proposed including both traditional forecasting models and deep learning models. Traditional forecasting models have limitations since they cannot describe the complex nonlinear relationship in climatic data, resulting in low forecasting accuracy. Deep learning techniques have promising non-linear processing capabilities in weather forecasting. To further advance the integration of deep learning in climate change forecasting, we have developed a hybrid model called CNN-ResNet50-LSTM, comprising a Convolutional Neural Network (CNN), a Deep Convolutional Network (ResNet50), and a Long Short-Term Memory (LSTM) model to predict two climate change factors: temperature and wind power. The experiment was conducted using three publicly available datasets: Wind Turbine Scada (Scada) Dataset, Saudi Arabia Weather history (SA) dataset, and Wind Power Generation Data for 4 locations (WPG) dataset. The forecasting accuracy is evaluated using several evaluation metrics, including the coefficient of determination (), Mean Squared Error (MSE), Mean Absolute Error (MAE), Median Absolute Error (MedAE) and Root Mean Squared Error (RMSE). The proposed CNN-ResNet50-LSTM model was also compared to five regression models: Dummy Regressor (DR), Kernel Ridge Regressor (KRR), Decision Tree Regressor (DTR), Extra Trees Regressor (ETR), and Stochastic Gradient Descent Regressor (SGDR). Findings revealed that CNN-ResNet50-LSTM model achieved the best performance, with scores of 98.84% for wind power forecasting in the Scada dataset, 99.01% for temperature forecasting in the SA dataset, 98.58% for temperature forecasting and 98.35% for wind power forecasting in the WPG dataset. The CNN-ResNet50-LSTM model demonstrated promising potential in forecasting both temperature and wind power. Additionally, we applied the CNN-ResNet50-LSTM model to predict climate changes up to 2030 using historical data, providing insights that highlight its potential for future forecasting and decision-making.

Abstract 7446: Stratification of cell therapies in solid tumor organoids using deep learning-derived imaging metrics

Abstract Background. High-throughput screening of immunotherapies is crucial for identifying promising candidates against various cancer indications and accelerating the pre-clinical development of cell therapies. Two technologies have shown promise in quantifying therapy effectiveness: patient-derived organoid (PDO) models co-cultured with cell therapies, and deep learning-based computer vision, which facilitates large-scale, automated image analysis to extract quantitative metrics of treatment efficacy and cell biology effects. Here, we use deep networks to predict PDO viability from brightfield images and compute explainable features which are used to stratify a large cohort of cell therapies, tested on multiple tumoroid lines and indications. Methods. Time-lapse confocal microscopy was used to record images of a large cohort of 27 cell therapies co-cultured with 15 different patient-derived tumor organoids (PDO) lines, across 8 cancer indications (breast, colorectal, endometrial, gastric, head and neck, liver, lung, pancreas) over the course of 72 hours, and at 2 different effector-to-target concentrations. PDO viability was measured by terminal TO-PRO-3 staining. A convolutional deep network was trained to perform label-free predictions of PDO viability from brightfield images at each timepoint (N=11, 110) and evaluated on a held-out test set (N=2, 112). Deep learning segmentation models were further used to co-localize tumoroid and immune cells, and extract a set of 6 explainable features - or spatial phenotypes - which quantified 1) tumoroid area, 2) cell apoptosis intensity and 3) temporal dynamics, as well as 4) immune proliferation, 5) immune infiltration and its 6) temporal dynamics. We fitted a linear model to measure their contribution in predicting terminal viability, and used hierarchical clustering to assess associations with exposure to different therapies. Results. Our bright-field model of viability was found to be highly concordant with ground truth viability from terminal vital dye stain on a held-out test set, representing 11 TO lines across 8 distinct cancer types (Pearson’s r=0.76). A linear model of spatial phenotypes showed high correlation with terminal viability (Pearson’s r=0.70), while clustering of spatial phenotypes stratified samples into groups of non-engineered and engineered therapies with and without an additional compound. Conclusion. Our study highlights that imaging-derived metrics from brightfield imaging can quantify treatment potency and cell phenotypes in ex-vivo screens. The approach pursued here is highly scalable and permits the analysis of large scale screening experiments to assess the impact of cell therapies on PDOs. Using explainable features further allows to stratify therapies into biologically meaningful groups based on the measured phenotypic readouts. Citation Format: Luca Lonini, Madhavi Kannan, Geoffrey Schau, Stanislaw Szydlo, Vignesh Krishnaraja, Sonal Khare, Daniel Rabe, Michael Streit, Mary Flaherty, Brandon Mapes, Mark Prytyskach, Stefan Kiesgen, Gurpanna Saggu, Megan McAfee, Kate Sasser, Justin Guinney, Richard Klinghoffer, Chi-Sing Ho. Stratification of cell therapies in solid tumor organoids using deep learning-derived imaging metrics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 7446.

A Modified Deep Convolutional Network for Detection of Covid19 from Chest X-Rays Based on Concatenation of Image Preprocessing Techniques and RESnCOV

The fast-spreading coronavirus disease called COVID-19 has impacted millions of people worldwide. It becomes difficult for medical experts to rapidly detect the illness and stop its spread because of its rapid growth and rising numbers. One of the newer areas of study where this issue can be more carefully addressed is medical image analysis. In this study, we implemented an image processing system utilizing deep learning and neural networks to previse the 2019-nCoV using chest roentgen ray images. In order to recognize COVID-19 positive and healthy patients using chest roentgen ray images, this paper suggests employing convolutional neural networks, deep learning, and machine learning. We proposed a neural network composed of various features taken from two convolutional neural networks, ResNet50 and ResNet152V2, in order to successfully manage the intricate structural complexity of an image. We tested our network on 7940 images to see how well it performs in real-world situations. The proposed network detects normal and COVID-19 cases with an average accuracy of 95% and can be used as an aid in the radiology department.

Crop Disease Detection System Using Convolutional Neural Network

One of the essential and tedious task in agricultural practices detecting of disease on crops. It requires huge time as well as skilled labour. This paper proposes a smart and efficient technique for the detection of crop disease which uses machine learning techniques. Every year India loses a significant amount of annual crop yield due to unidentified plant diseases. The traditional method of disease detection is manual examination by either farmers or experts, which may be time-consuming and inaccurate. It is proving infeasible for many small and medium-sized farms around the world. To mitigate this issue, a computer aided disease recognition model is proposed. It uses leaf image classification with the help of deep convolutional networks. In this paper, CNN was proposed to detect plant disease. It has three processing steps namely feature extraction, downsizing image, and classification. In CNN, the convolutional layer extracts the feature from the plant image. It helps to give personalized recommendations to the farmers based on soil features, temperature, and humidity

Automated segmentation of hyperreflective foci in OCT images for diabetic retinopathy using deep convolutional networks

Optical coherence tomography (OCT) is being investigated in diabetic retinopathy (DR) diagnostics as a real-time evaluation tool. Currently, OCT images are the main methods for the diagnosis of patients with DR. Hyperreflective foci (HRF) are potential biomarkers for the diagnosis and prediction of the progression and prognosis of patients with DR. The development of artificial intelligence (AI) models for segmenting HRF is of great significance for the clinical diagnosis and treatment of patients with DR. The purpose of this study is to construct a deep-learning algorithm that automatically segments the HRF in OCT images, helping ophthalmologists make early diagnosis and evaluate the prognosis of patients with DR. In this paper, to investigate the algorithms that are appropriate for the segmentation of HRF, we propose an HRF segmentation algorithm on the basis of Attention U-Net. We fuse the features of each layer and use the fused multi-scale information to guide the generation of the attention map. Then, we embed a hybrid attention module of space and channel at the decoder end of the network to capture the spatial and channel correlations of the feature map, making the network focus on the location and channels related to the target region. We propose a novel algorithm, to our knowledge, based on Attention U-Net and the experimental results on 172 OCT images from 50 patients with DR demonstrated that our method is effective for the HRF segmentation. In five-fold cross-validation, the dice similarity coefficient (DSC), sensitivity (SE), and precision (P) reach 63.79±0.94, 66.66±2.54, and 67.10±1.96, respectively. The overall segmentation effect of this model surpasses that of the other four networks, and the HRF can be segmented more accurately and identified more easily. In a segment model, balancing SE and P is difficult. We developed an improved Attention U-Net that effectively segments HRF with high SE and P, outperforming other algorithms in HRF segmentation. This model holds significant potential for the early detection, treatment evaluation, and prognosis assessment of patients with diabetic retinopathy (DR).

Hybrid Quantum Deep Convolutional Generative Adversarial Networks for Channel Prediction and Performance Enhancement in Large‐Scale MIMO‐OFDM Systems

ABSTRACTMultiple‐output orthogonal frequency division multiplexing (MIMO‐OFDM) systems demand accurate channel prediction for optimal performance. This research presents an innovative approach employing a hybrid quantum deep convolutional generative adversarial network (HQDCGAN) to enhance channel prediction, minimize error vector magnitude, reduce peak power, and mitigate adjacent channel leakage ratio (HQDCGAN‐MIMO‐OFDM) proposed. This approach implements a peak‐to‐average power ratio (PAPR) reduction module utilizing HQDCGAN trained with lower PAPR data acquired through simplified clipping with filtering (SCF) technique. The proposed HQDCGAN architecture leverages pyramidal dilated convolutions and attention mechanisms to extract multi‐scale features from OFDM channel data. By incorporating attention mechanisms, the model dynamically focuses on crucial information, refining the channel prediction process. The primary objective is to exploit the network's capability to learn complex spatial–temporal correlations within OFDM channel signals. These strategy goals are to significantly improve the accuracy and, robustness of channel prediction, leading to minimized error vector magnitude (EVM) and mitigated issues related to peak power and adjacent channel leakage ratio (ACLR). To validate the efficiency of the proposed HQDCGAN‐MIMO‐OFDM the evaluation metrics such as spectral efficiency, peak‐to‐average power ratio, BER, SNR, and throughput are quantitatively analyzed. The proposed method CP‐LSMIMO‐OFDM‐HQDCGAN gives 20.67%, 12.78%, and 19.56% low bit error rate, 21.66%, 23.09%, and 25.11% low reduction in PAPR and 23.76%, 30.45% and 18.97% high throughput with existing methods like TOP‐ADMM, RNN‐DNN‐MIMO‐OFDM, and IA‐MIMO‐OFDM methods, respectively.

A tractor transmission systems vibration fault data augmentation and diagnosis method based on DCGAN

The actual production process is often faced with insufficient fault data. Traditional data augmentation algorithms are prone to problems such as overfitting and pattern collapse. To address the issues, a tractor transmission systems vibration fault data augmentation and diagnosis method based on deep convolutional generative adversarial network is proposed. First, the original vibration signal is converted to a time–frequency diagram, which serves as the input to the generative adversarial network. Subsequently, the two timescale update rule strategy and gradient penalty are applied to stabilize the training process of the generative adversarial network, and the self-attention mechanism is introduced to enhance the feature extraction capability of the generative adversarial network. Furthermore, an enhanced AlexNet based on the convolutional block attention module is developed to improve the diagnostic performance. The approach was verified using datasets from Case Western Reserve University (CWRU), Huazhong University of Science and Technology (HUST), and a laboratory gearbox dataset. Results show 99% diagnostic accuracy for CWRU, over 97.5% for the laboratory dataset, and 98% accuracy under a single condition for the HUST gear dataset, with accuracy exceeding 95.3% under other conditions. This indicates that the method improves diagnostic precision and model generalization under data scarcity.

Two-Stage Video Violence Detection Framework Using GMFlow and CBAM-Enhanced ResNet3D

Video violence detection has gained significant attention in recent years due to its applications in surveillance and security. This paper proposes a two-stage framework for detecting violent actions in video sequences. The first stage leverages GMFlow, a pre-trained optical flow network, to capture the temporal motion between consecutive frames, effectively encoding motion dynamics. In the second stage, we integrate these optical flow images with RGB frames and feed them into a CBAM-enhanced ResNet3D network to capture complementary spatiotemporal features. The attention mechanism provided by CBAM enables the network to focus on the most relevant regions in the frames, improving the detection of violent actions. We evaluate the proposed framework on three widely used datasets: Hockey Fight, Crowd Violence, and UBI-Fight. Our experimental results demonstrate superior performance compared to several state-of-the-art methods, achieving AUC scores of 0.963 on UBI-Fight and accuracies of 97.5% and 94.0% on Hockey Fight and Crowd Violence, respectively. The proposed approach effectively combines GMFlow-generated optical flow with deep 3D convolutional networks, providing robust and efficient detection of violence in videos.

Deep spatio-temporal dependent convolutional LSTM network for traffic flow prediction

With the rapid development of economy, the concept of intelligent transportation system (ITS) and smart city has been mentioned. The most important part of building them is whether they can accurately predict traffic flow. An accurate traffic flow forecast can help manage traffic, plan travel paths in advance, and rationally allocate public resources such as shared bicycles. The biggest difficulty in this task is how to solve the problem of spatial imbalance and the problem of temporal imbalance. In this paper, we propose a deep learning algorithm STDConvLSTM. Firstly, for spatial features, most scholars use convolutional neural networks (with fixed kernel size) to capture. However, this does not solve the problem of spatial imbalance, i.e. each region has a different size of correlated regions (e.g., the busy area has a wider range of correlated regions). In this paper, we design a space-dependent attention mechanism, which assigns a convolutional neural network with a different kernel size to each region through attention weights. Secondly, for time features, most scholars use time series prediction models, such as recurrent neural networks and their variants. However, in the actual forecasting process, the importance of historical data in different time steps is not the same. In this paper, we design a time-dependent attention mechanism that assigns different weights to historical data to solve the time imbalance. In the end, we ran experiments on two real-world data sets and achieve good performance.

Parkinson's disease tremor prediction towards real-time suppression: A self-attention deep temporal convolutional network approach.

Accurate prediction of Parkinson's disease tremor (PDT) is crucial for developing assistive technologies; however, this is challenging due to the nonlinear, stochastic, and nonstationary characteristics of PDT, which substantially vary among patients and their activities. Moreover, most models only have one-step prediction capabilities, which causes delays in real-time applications. This paper proposes a self-attention deep temporal convolutional network (SADTCN) model for the real-time prediction of hand-arm PDT signals from different activities and joint angular motions. The SADTCN can capture both short- and long-term dependencies and complex temporal and spatial dynamics of PDT signals and hence, can effectively adapt to varying tremor characteristics. The performance of the proposed model is evaluated using experimental hand-arm PDT data. The results show that the SADTCN outperforms existing deep learning (DL) models by accurately predicting varying tremor amplitudes and frequencies multi-step ahead. Moreover, we performed spectrum analysis on the measured and predicted signal using the short-time Fourier transform (STFT) as a measure of potential active tremor control and found that SADTCN can accurately determine the transience of tremor amplitude in frequency and time. Finally, we run the Wilcoxon signed-rank statistical test and the results show a statistically significant improvement in the proposed model over the other DL models in all conditions. Therefore, the SADTCN can overcome the nonstationary, nonlinear, and stochastic nature of PDT to perform multi-step prediction with high accuracy, robustness, and generalizability in unseen testing data.

Automated tomato leaf disease recognition using deep convolutional networks

Agriculture is essential for the entire global population. An advanced, robust, and empirically sound agriculture sector is essential for nourishing the global population. Various leaf diseases cause financial hardships for farmers and related businesses. Early identification of foliar diseases in crops would greatly help farmers, leading to a substantial increase in agricultural productivity. The tomato is a widely recognized and nourishing food that is easily accessible and highly favored by farmers. Early diagnosis of tomato leaf diseases is crucial to maximize tomato crop production. This study aims to utilize a deep learning approach to accurately detect and classify damaged leaves and disease patterns in tomato leaf images. By employing a substantial quantity of deep convolutional network models, we achieved a high level of precision in diagnosing the condition. The dataset used in our study work is a self-contained dataset obtained by direct observation of tomato fields in rural areas of Bangladesh. It consists of four classes: healthy, black mold, grey mold, and powdery mildew. In this study work, we utilized various image pre-processing techniques and applied VGG16, InceptionV3, DenseNet121, and AlexNet models. Our results showed that the DenseNet121 model attained the higher accuracy of 97%. This discovery guarantees accurate detection of tomato diseases in a rapid manner, ushering in a new agricultural revolution.

Regional Image Quality Scoring for 2-D Echocardiography Using Deep Learning.

Regional Image Quality Scoring for 2-D Echocardiography Using Deep Learning.

IoT security using heuristic aided symmetric convolution-based deep temporal convolution network for intrusion detection by extracting multi-cascaded deep attention features

IoT security using heuristic aided symmetric convolution-based deep temporal convolution network for intrusion detection by extracting multi-cascaded deep attention features