Dense Convolutional Neural Network Research Articles

Color correction method of interior decoration engineering based on dense convolution neural network

Color correction method of interior decoration engineering based on dense convolution neural network

Landslide Detection Mapping Employing CNN, ResNet, and DenseNet in the Three Gorges Reservoir, China

Landslide detection mapping (LDM) is the basis of the field of landslide disaster prevention; however, it has faced certain difficulties. The Three Gorges Reservoir area of the Yangtze River has been one of the most intensively evaluated areas for landslide prevention in the world, due to the high frequency of landslide disasters here. In this article, we constructed an accurate LDM model based on convolutional neural networks, residual neural networks, and dense convolutional neural networks (DenseNets) that considers “ZY-3” high spatial resolution (HSR) data and conditioning factors (CFs). In this article, 19 factors based on remote sensing (RS) images, topographical and geological data associated with historical landslide locations were randomly divided into training (70% of total) and testing (30%) datasets. The experimental results show that the accuracy (ACC) of these three LDM models is above 0.95, indicating that the deep neural networks aimed at landslide detection performed well. Furthermore, DenseNet with RS images and CFs can accurately detect landslides. Specifically, DenseNet with RS images and CFs outperforms the other five models by considering the evaluation metrics, which exhibited Kappa coefficient improvements of 0.01–0.04 and ACC improvements of 0.02–0.3%. Among all the factors, elevation factor has a high importance of 0.727, which is the most important factors found in this landslide model construction experiment.

Screening of COVID-19 Suspected Subjects Using Multi-Crossover Genetic Algorithm Based Dense Convolutional Neural Network

Fast and accurate screening of novel coronavirus (COVID-19) suspected subjects plays a vital role in timely quarantine and medical care. Deep transfer learning-based screening models on chest X-ray (CXR) are effective for countering the COVID-19 outbreak. However, an efficient screening of COVID-19 is still a huge task due to the spatial complexity of CXRs. In this paper, a dense convolutional neural network (DCov-Net) based transfer learning model is proposed for the screening of COVID-19 suspected subjects using CXR images. A modified multi-crossover genetic algorithm (MMCGA) is then proposed to tune the hyper-parameters of DCov-Net. Majority of the existing COVID-19 diagnosis models are not interpretable as they do not provide any transparency to the users. Therefore, the concept of heat-maps is used to achieve explainability and interpretability. MMCGA based DCov-Net is implemented on a multiclass dataset that contains four different classes. Experimental results reveal that MMCGA based DCov-Net achieves better performance than the existing models. The proposed MMCGA based DCov-Net can be utilized for initial screening of COVID-19 suspected subjects with an accuracy of 99.34 ± 0.51 %.

Deep convolutional neural networks for multiplanar lung nodule detection: Improvement in small nodule identification.

PurposeEarly detection of lung cancer is of importance since it can increase patients’ chances of survival. To detect nodules accurately during screening, radiologists would commonly take the axial, coronal, and sagittal planes into account, rather than solely the axial plane in clinical evaluation. Inspired by clinical work, the paper aims to develop an accurate deep learning framework for nodule detection by a combination of multiple planes.MethodsThe nodule detection system is designed in two stages, multiplanar nodule candidate detection, multiscale false positive (FP) reduction. At the first stage, a deeply supervised encoder–decoder network is trained by axial, coronal, and sagittal slices for the candidate detection task. All possible nodule candidates from the three different planes are merged. To further refine results, a three‐dimensional multiscale dense convolutional neural network that extracts multiscale contextual information is applied to remove non‐nodules. In the public LIDC‐IDRI dataset, 888 computed tomography scans with 1186 nodules accepted by at least three of four radiologists are selected to train and evaluate our proposed system via a tenfold cross‐validation scheme. The free‐response receiver operating characteristic curve is used for performance assessment.ResultsThe proposed system achieves a sensitivity of 94.2% with 1.0 FP/scan and a sensitivity of 96.0% with 2.0 FPs/scan. Although it is difficult to detect small nodules (i.e., <6 mm), our designed CAD system reaches a sensitivity of 93.4% (95.0%) of these small nodules at an overall FP rate of 1.0 (2.0) FPs/scan. At the nodule candidate detection stage, results show that the system with a multiplanar method is capable to detect more nodules compared to using a single plane.ConclusionOur approach achieves good performance not only for small nodules but also for large lesions on this dataset. This demonstrates the effectiveness of our developed CAD system for lung nodule detection.

RSNN: A Software/Hardware Co-Optimized Framework for Sparse Convolutional Neural Networks on FPGAs

Convolutional Neural Networks (CNNs) have been shown to be very useful in image recognition and other Artificial Intelligence (AI) applications, however, at the expense of intensive computation requirement. To address the challenge of overwhelming calculation requirements, researchers have proposed various network pruning techniques. But, due to the irregular sparse patterns, unstructured sparse networks are difficult to compute efficiently on either Graphic processing units (GPUs) or Field Programmable Gate Arrays (FPGAs). In this paper, we propose a software/hardware co-optimized Reconfigurable Sparse convolutional Neural Network accelerator design (RSNN) on FPGAs. A novel sparse convolution dataflow is proposed with simpler control logic than existing mux-based selection logic. To balance the computation load on different Processing Units (PUs), we propose a software-based load-balance aware pruning technique as well as a kernel merging method. Experimental results show that RSNN is $2.41\times - 7.91\times $ better on Digital Signal Processor (DSP) efficiency than previous dense CNN FPGA accelerators, and $1.23\times - 2.93\times $ better than previous sparse CNN FPGA accelerators.

Brain MRI artefact detection and correction using convolutional neural networks

Background and Objective: Brain MRI is one of the most commonly used diagnostic imaging tools to detect neurodegenerative disease. Diagnostic image quality is a key factor to enable robust image analysis algorithms developed for downstream tasks such as segmentation. In clinical practice, one of the main challenges is the presence of image artefacts, which can lead to low diagnostic image quality.Methods: In this paper, we propose using dense convolutional neural networks to detect and a residual U-net architecture to correct motion related brain MRI artefacts. We first generate synthetic artefacts using an MR physics based corruption strategy. Then, we use a detection strategy based on dense convolutional neural network to detect artefacts. The detected artefacts are corrected using a residual U-net network trained on corrupted data.Results: Our pipeline for detection and correction of artefacts is capable of reaching not only better quality image quality, but also better segmentation accuracy of stroke segmentation. The algorithm is validated using 28 cases brain MRI stroke segmentation dataset and showed an accuracy of 97.8% for detecting artefacts in our experiments. We also illustrated the improved image quality and segmentation accuracy with the proposed correction algorithm.Conclusions: Ensuring high image quality and high segmentation quality jointly can improve the automatic image analysis pipelines and reduce the influence of low image quality on final prognosis. With this work, we illustrate a performance analysis on brain MRI stroke segmentation.

IoMT-Based Automated Detection and Classification of Leukemia Using Deep Learning

For the last few years, computer-aided diagnosis (CAD) has been increasing rapidly. Numerous machine learning algorithms have been developed to identify different diseases, e.g., leukemia. Leukemia is a white blood cells- (WBC-) related illness affecting the bone marrow and/or blood. A quick, safe, and accurate early-stage diagnosis of leukemia plays a key role in curing and saving patients' lives. Based on developments, leukemia consists of two primary forms, i.e., acute and chronic leukemia. Each form can be subcategorized as myeloid and lymphoid. There are, therefore, four leukemia subtypes. Various approaches have been developed to identify leukemia with respect to its subtypes. However, in terms of effectiveness, learning process, and performance, these methods require improvements. This study provides an Internet of Medical Things- (IoMT-) based framework to enhance and provide a quick and safe identification of leukemia. In the proposed IoMT system, with the help of cloud computing, clinical gadgets are linked to network resources. The system allows real-time coordination for testing, diagnosis, and treatment of leukemia among patients and healthcare professionals, which may save both time and efforts of patients and clinicians. Moreover, the presented framework is also helpful for resolving the problems of patients with critical condition in pandemics such as COVID-19. The methods used for the identification of leukemia subtypes in the suggested framework are Dense Convolutional Neural Network (DenseNet-121) and Residual Convolutional Neural Network (ResNet-34). Two publicly available datasets for leukemia, i.e., ALL-IDB and ASH image bank, are used in this study. The results demonstrated that the suggested models supersede the other well-known machine learning algorithms used for healthy-versus-leukemia-subtypes identification.

Underwater Acoustic Target Classification Based on Dense Convolutional Neural Network

In oceanic remote sensing operations, underwater acoustic target recognition is always a difficult and extremely important task of sonar systems, especially in the condition of complex sound wave propagation characteristics. The expensively learning recognition model for big data analysis is typically an obstacle for most traditional machine learning (ML) algorithms, whereas the convolutional neural network (CNN), a type of deep neural network, can automatically extract features for accurate classification. In this study, we propose an approach using a dense CNN model for underwater target recognition. The network architecture is designed to cleverly reuse all former feature maps to optimize classification rates under various impaired conditions while satisfying low computational cost. In addition, instead of using time–frequency spectrogram images, the proposed scheme allows directly utilizing the original audio signal in the time domain as the network input data. Based on the experimental results evaluated on the real-world data set of passive sonar, our classification model achieves the overall accuracy of 98.85% at 0-dB signal-to-noise ratio (SNR) and outperforms traditional ML techniques, as well as other state-of-the-art CNN models.

Texture features from computed tomography correlate with markers of severity in acute alcohol-associated hepatitis

The aim of this study was to use texture analysis to establish quantitative CT-based imaging features to predict clinical severity in patients with acute alcohol-associated hepatitis (AAH). A secondary aim was to compare the performance of texture analysis to deep learning. In this study, mathematical texture features were extracted from CT slices of the liver for 34 patients with a diagnosis of AAH and 35 control patients. Recursive feature elimination using random forest (RFE-RF) was used to identify the best combination of features to distinguish AAH from controls. These features were subsequently used as predictors to determine associated clinical values. To compare machine learning with deep learning approaches, a 2D dense convolutional neural network (CNN) was implemented and trained for the classification task of AAH. RFE-RF identified 23 top features used to classify AAH images, and the subsequent model demonstrated an accuracy of 82.4% in the test set. The deep learning CNN demonstrated an accuracy of 70% in the test set. We show that texture features of the liver are unique in AAH and are candidate quantitative biomarkers that can be used in prospective studies to predict the severity and outcomes of patients with AAH.

Image super resolution based on residual dense CNN and guided filters

Convolutional neural networks (CNNs) have recently made impressive results for image super-resolution (SR). Our goal is to introduce a new image SR framework rely on a CNN. In this paper, the input image is decomposed into luminance channel and chromatic channels. A designed network based on a residual dense network is introduced to extract the hierarchical features from luminance part. The bicubic interpolation is simply used to upscale low resolution (LR) chromatic channels. However, this step degrades the chromatic channels. To tackle this issue, the SR reconstructed luminance channel is applied as the reference image in guided filters to promote the interpolated chromatic channels. Guided filters technique has ability to retain sharp edges and fine details from the reference image and carry them to the target images. Extensive experiments on several commonly used image SR testing datasets demonstrate that our framework has the ability to extract features and outperforms existing well-known techniques for image SR by LR image into the high resolution (HR) image efficiently.

Towards an accurate radar waveform recognition algorithm based on dense CNN

Existing algorithms for radar waveform classification currently exhibit the lower recognition accuracy, especially at the lower signal to noise ratio (SNR) environment. To remedy these flaws, this paper proposes an accurate automatic modulation classification algorithm based on dense convolutional neural networks (AAMC-DCNN). The algorithm owns the competitive advantages of strengthening the feature reuse and extracting the detailed feature, for improving the recognition performance of radar waveform at the lower SNR. First, the dense convolutional neural networks (CNN) are designed, which connects each layer to every other layer in a feed-forward pattern. In the latter, 8 types of signals are converted into time-frequency images by choi-williams distribution (CWD), and the large training and testing datasets are fabricated. Then, the transfer learning and Adam optimization are introduced. Finally, the experimental analyses are carried out to evaluate the recognition performance. It is worth mentioning that the classification accuracy can be up to 93.4% when the SNR is −8 dB, and even reach to 100% at 0 dB, which demonstrates the superior performance over others. The present work provides a sound experimental basis for further studying automatic modulation classification for the sake of future field application in electronic warfare systems.

An Efficient Hardware Accelerator for Structured Sparse Convolutional Neural Networks on FPGAs

Deep convolutional neural networks (CNNs) have achieved state-of-the-art performance in a wide range of applications. However, deeper CNN models, which are usually computation consuming, are widely required for complex artificial intelligence (AI) tasks. Though recent research progress on network compression, such as pruning, has emerged as a promising direction to mitigate computational burden, existing accelerators are still prevented from completely utilizing the benefits of leveraging sparsity due to the irregularity caused by pruning. On the other hand, field-programmable gate arrays (FPGAs) have been regarded as a promising hardware platform for CNN inference acceleration. However, most existing FPGA accelerators focus on dense CNN and cannot address the irregularity problem. In this article, we propose a sparsewise dataflow to skip the cycles of processing multiply-and-accumulates (MACs) with zero weights and exploit data statistics to minimize energy through zeros gating to avoid unnecessary computations. The proposed sparsewise dataflow leads to a low bandwidth requirement and high data sharing. Then, we design an FPGA accelerator containing a vector generator module (VGM) that can match the index between sparse weights and input activations according to the proposed dataflow. Experimental results demonstrate that our implementation can achieve 987-, 46-, and 57-imag/s performance for AlexNet, VGG-16, and ResNet-50 on Xilinx ZCU102, respectively, which provides $1.5\times $ – $6.7\times $ speedup and $2.0\times $ – $6.0\times $ energy efficiency over previous CNN FPGA accelerators.

EMAP-DCNN: A NOVEL MATHEMATICAL MORPHOLOGY AND DEEP LEARNING COMBINED FRAMEWORK FOR HYPERSPECTRAL IMAGE CLASSIFICATION

Abstract. The classification of hyperspectral image (HSI) with high spectral and spatial resolution represents an important and challenging task in image processing and remote sensing (RS) domains due to the problem of computational complexity and big dimensionality of the remote sensing images. The spatial and spectral pixel characteristics have crucial significance for hyperspectral image classification and to take into account these two types of characteristics, various classification and feature extraction methods have been developed to improve spectral-spatial classification of remote sensing images for thematic mapping purposes such as agricultural mapping, urban mapping, emergency mapping in case of natural disasters... In recent years, mathematical morphology and deep learning (DL) have been recognized as prominent feature extraction techniques that led to remarkable spectral-spatial classification performances. Among them, Extended Multi-Attribute Profiles (EMAP) and Dense Convolutional Neural Network (DCNN) are considered as robust and powerful approaches such as the work in this paper is based on these two techniques for the feature extraction stage and used in two combined manners and constructing the EMAP-DCNN frame. The experiments were conducted on two popular datasets: “Indian Pines” and “Huston” hyperspectral datasets. Experimental results demonstrate that the two proposed approaches of the EMAP-DCNN frame denoted EMAP-DCNN 1, EMAP-DCNN 2 provide competitive performances compared with some state-of-the-art spectral-spatial classification methods based on deep learning.

A Novel OFDM Autoencoder Featuring CNN-Based Channel Estimation for Internet of Vessels

This article proposes a novel orthogonal frequency-division multiplexing (OFDM) autoencoder featuring convolutional neural networks (CNNs)-based channel estimation for marine communications with complex and fast-changing environments. We demonstrate that the proposed OFDM autoencoder system can be generalized to work under various channel environments, different throughputs, while outperforming the traditional OFDM counterparts, especially when working at high throughputs. In addition, since OFDM systems require accurate channel estimations to function properly, this treatise also proposes a new channel estimation algorithm for OFDM systems that combine the power of deep learning (DL) with the philosophy of super-resolution reconstruction, which uses dense convolutional neural networks (Dense-Nets) to reconstruct low-resolution pilot information images into high-resolution full-channel impulse responses (CIRs). The Dense-Net structure has the characteristics of dense connections and feature multiplexing. The simulation results show that under slow fading, the proposed channel estimator (CE) can estimate the CIRs perfectly. Under fast fading, the proposed CE outperforms the existing learning-based algorithms with fewer neural network parameters. Therefore, the proposed novel autoencoder scheme and the powerful CE are potentially attractive approaches for the Internet of Vessels (IoV).

WITHDRAWN: Residual Dense Convolutional Neural Network for Image Super-Resolution

Abstract Image super-resolution (SR) is the resource-intensive process of scaling low-resolution (LR) images to high resolution (HR). There are a number of difficulties with SR and related areas of computer vision, so it is crucial that we develop effective SR techniques. The evolution of deep learning technology, especially convolutional neural networks (CNNs), has led to a number of approaches to image improvement, including image SR. The aim of this paper is to develop a fast image SR technique based on a CNN. We introduce a tripartite CNN for multi-color SR enhancement. First, we develop a deconvolution step at the end of the network structure, and mapping is then applied to the input LR luminance, which is used as a reference for the HR image. Second, a novel residual dense network is developed to exploit the hierarchical features of the convolutional layers. This step merges different local features using densely connected convolutional layers. Third, we apply guided filters to improve the chromatic interpolation. Guided filters identify sharp edges and fine details in the reference image and reproduce them in the output images. Finally, we extract further mapping layers using smaller filters. Our experimental results confirm that this algorithm achieves visual and quantitative results superior to other state-of-the-art methods while maintaining computational efficiency.

An optimized dense convolutional neural network model for disease recognition and classification in corn leaf

An optimized dense convolutional neural network (CNN) architecture (DenseNet) for corn leaf disease recognition and classification is proposed in this paper. Corn is one of the most cultivated grain throughout the world. Corn crops are highly susceptible to certain leaf diseases such as corn common rust, corn gray leaf spot, and northern corn leaf blight are very common. Symptoms of these leaf diseases are not differentiable in their nascent stages. Hence, the current research presents a solution through deep learning so that crop health can be monitored and, it will lead to an increase in the quantity as well as the quality of crop production. The proposed optimized DenseNet model has achieved an accuracy of 98.06%. Besides, it uses significantly lesser parameters as compared to the various existing CNN such as EfficientNet, VGG19Net, NASNet, and Xception Net. The performance of the optimized DenseNet model has been contrasted with the current CNN architectures by considering two (time and accuracy) quality measures. This study indicates that the performance of the optimized DenseNet model is close to that of the established CNN architectures with far fewer parameters and computation time.

102-OR: Detection of Diabetes from Whole-Body Magnetic Resonance Imaging Using Deep Learning

Obesity is one of the main drivers of the globally rising prevalence of type 2 diabetes (T2D). Yet, obesity is not uniformly associated with metabolic consequences. The location of fat accumulation is critical for metabolic health. Specific patterns of body fat distribution, such as an increased ratio of visceral to subcutaneous fat, are closely related to insulin resistance which is crucial in the pathogenesis of T2D. There might be further, hitherto unknown features of body fat distribution which could additionally contribute to the disease. We used a machine learning approach with dense convolutional neural networks (DCNN) to detect diabetes related variables from 2371 T1-weighted whole-body magnetic resonance image (MRI) data sets. Each single measurement was labelled by sex, age, BMI, insulin sensitivity, HbA1c and prediabetes or incident diabetes. The result was compared to conventional models using segmented body fat compartment volumes. Anatomical labels were assigned to locations in the DCNN gradient heatmaps that are critical for discrimination. The AUC-ROC was 0.87 for the discrimination of diabetes and 0.68 for prediabetes. Classification performance was superior to conventional models. Mean absolute regression errors were comparable to those of the conventional models. Heatmaps clearly showed that lower visceral abdominal regions were most critical in diabetes classification, while other significant areas comprised upper legs, arms and the neck region.Our results show that diabetes is detectable from whole-body MRI without any blood glucose measurement. Our technique of heatmap visualization unravels plausible anatomical regions and highlights the leading role of fat accumulation in the lower abdomen in the pathogenesis of T2D. Disclosure R. Wagner: Advisory Panel; Self; Novo Nordisk A/S. Speaker’s Bureau; Self; Novo Nordisk A/S. Other Relationship; Self; Eli Lilly and Company. B. Dietz: None. J. Machann: None. P. Schwab: Employee; Self; Roche Pharma. J.K. Dienes: Advisory Panel; Spouse/Partner; Novo Nordisk A/S. Speaker’s Bureau; Spouse/Partner; Novo Nordisk A/S. Other Relationship; Spouse/Partner; Eli Lilly and Company. S. Reichert: Other Relationship; Self; Lilly Diabetes. A.L. Birkenfeld: None. H. Haering: None. F. Schick: None. N. Stefan: None. M. Heni: Research Support; Self; Boehringer Ingelheim Pharmaceuticals, Inc., Sanofi. Speaker’s Bureau; Self; Novo Nordisk A/S. H. Preissl: None. B. Schölkopf: None. S. Bauer: None. A. Fritsche: None. Funding German Federal Ministry of Education and Research (01GI0925)

Fault detection and identification of rolling element bearings with Attentive Dense CNN

Deep learning (DL) applications have redefined the state of the art performance for bearing data driven fault detection and identification, a crucial demand of modern industrial systems. The success of DL methods, in the field of automatic fault diagnosis, is based on the usage of raw sensor data, contrary to conventional machine learning (ML) approaches in which manual extraction of features, from prior expertise knowledge, is necessary. However, DL approaches require a large amount of training data samples to be effective and to outperform competitive ML methods. In this study, we overcome this drawback by proposing the Attentive Dense Convolutional Neural Network (ADCNN), a DL network, which considers the temporal coherence of the data samples, by the combination of Dense Convolutional blocks with an attention mechanism. The proposed neural scheme has fewer unknown learning parameters and achieves accurate results with less training data as it appears from simulation cases on two famous rolling bearings fault detection benchmarks.

Learning from irregularly sampled data for endomicroscopy super-resolution: a comparative study of sparse and dense approaches

PurposeProbe-based confocal laser endomicroscopy (pCLE) enables performing an optical biopsy via a probe. pCLE probes consist of multiple optical fibres arranged in a bundle, which taken together generate signals in an irregularly sampled pattern. Current pCLE reconstruction is based on interpolating irregular signals onto an over-sampled Cartesian grid, using a naive linear interpolation. It was shown that convolutional neural networks (CNNs) could improve pCLE image quality. Yet classical CNNs may be suboptimal in regard to irregular data.MethodsWe compare pCLE reconstruction and super-resolution (SR) methods taking irregularly sampled or reconstructed pCLE images as input. We also propose to embed a Nadaraya–Watson (NW) kernel regression into the CNN framework as a novel trainable CNN layer. We design deep learning architectures allowing for reconstructing high-quality pCLE images directly from the irregularly sampled input data. We created synthetic sparse pCLE images to evaluate our methodology.ResultsThe results were validated through an image quality assessment based on a combination of the following metrics: peak signal-to-noise ratio and the structural similarity index. Our analysis indicates that both dense and sparse CNNs outperform the reconstruction method currently used in the clinic.ConclusionThe main contributions of our study are a comparison of sparse and dense approach in pCLE image reconstruction. We also implement trainable generalised NW kernel regression as a novel sparse approach. We also generated synthetic data for training pCLE SR.

Dense Contour-Imbalance Aware framework for Colon Gland Instance Segmentation

Colon gland images contain rich information about the patient’s condition and the automated instance segmentation results have been widely used to relieve the workload of pathologists and further assist pathologists in making clinical decisions. Previous methods utilize several deep layers’ features and auxiliary contour prediction results to model this task. However, they always fail to capture the complex representation of gland images and bring the contour imbalance problem, resulting in limited performance. To overcome these problems, in this paper, we propose a dense contour-imbalance aware (DCIA) framework by leveraging the recent advances in dense convolutional neural network (DenseNet) and focal loss (FL). First, we integrate all the semantic features generated by DenseNet to explore the “optimal” representation of input image. On this basis, to mitigate the contour imbalance problem in the training stage, we replace common cross-entropy objective with FL. Also, two post-processing techniques, i.e., morphology and convolutional conditional random fields (ConvCRFs) are adopted to further refine the predicted confidence maps, aiming to achieve better performance. Experimental results demonstrate that the proposal outperforms other popular methods on visual perception. Also, compared with the recent popular method DCAN, our framework has higher scores on the testing sets B in terms of three quantitative metrics Object F1, Object Dice, and Object Hausdorff with the improvement of 5.54%, 1.37% and 12.16%, respectively.