Dense Block Research Articles

Enhancing the thermal stability of S/Se based thermoelectric materials using self-generated oxide films

This study focuses on the thermal stability of Cu1.8S materials. During the ball milling process, metal powder is added to the milled Cu1.8S powder, and then the obtained powder is sintered using current assisted sintering to obtain dense blocks. The aim is to enable the added metal elements to spontaneously grow an oxide film on the surface of the material block at high temperature, achieving protection of the material matrix. The thermal stability of materials is evaluated by utilizing changes in room temperature phase composition before and after high-temperature heat treatment, changes in material electrical conductivity during high-temperature processes, and cyclic electrical transmission performance testing. By observing the surface oxide film state of the material after high-temperature treatment, the commonalities and differences in the effects of different element additions on the thermal stability of the material were analyzed. The effects of different metal elements on material hardness and electrical transmission performance were evaluated. It is found that adding metal powder can effectively improve the thermal stability of Cu1.8S, improve material hardness, and regulate the electrical transmission performance of the material. The characteristics of the oxide film formed by the spontaneous growth of metal elements and oxygen on the surface of the material substrate determine the effectiveness of the oxide film in protecting the material from high temperatures. The pure Cu1.8S bulk can only maintain stability at 300 oC, and the addition of Cr, Al, 316L, Fe, and Mn powder respectively increased the stable temperature of the material to 400, 400, 450, 450, 500 oC. Adding metal elements to the material matrix to grow an oxide film on the surface of the material to prevent high-temperature oxidation or decomposition is an effective way to improve the thermal stability of S/Se compounds.

Spatial Health Predictors for Depressive Disorder in Manhattan: A 2020 Analysis.

Background Urban cores often present extreme disparities in the distribution of wealth and income. They also vary in health outcomes, especially regarding mental welfare. Dense urban blocks agglomerate many residents of various backgrounds, and extreme differences in income, commerce, and health may lead to variations in depressive disorder outcomes. More research is needed on public health characteristics that may affect depression in dense urban centers. Methods Data on 2020 public health characteristics for Manhattan Island was collected using the Centers for Disease Control's (CDC's) PLACES project. All Manhattan census tracts were used as the spatial observations, resulting in [Formula: see text]observations. A cross-sectional generalized linear regression (GLR) was used to fit a geographically weighted spatial regression (GWR), with tract depression rates as the endogenous variable. Data on the following eight exogenous parameters were incorporated: the percentage without health insurance, the percentage of those who binge drink, the percentage who receive an annual doctor's checkup, the percentage of those who are physically inactive, the percentage of those who experience frequent mental distress, the percentage of those who receive less than 7 hours of sleep each night, the percentage of those who report regular smoking, and the percentage of those who are obese. A Getis-Ord Gi* model was built to locate hot and cold spot clusters for depression incidence and an Anselin Local Moran's I spatial autocorrelation analysis was undertaken to determine neighborhood relationships between tracts. Results Depression hot spot clusters at the 90%-99% confidence interval (CI) were identified in Upper Manhattan and Lower Manhattan using the Getis-Ord Gi* statistic and spatial autocorrelation. Cold spot clusters at the 90%-99% CI were in central Manhattan and the southern edge of Manhattan Island. For the GLR-GWR model, only the lack of health insurance and mental distress variables were significant at the 95% CI, with an adjusted R-2 of 0.56. Noticeable inversions were observed in the spatial distribution of the exogenous coefficients across Manhattan, with a higher lack of insurance coefficients observed in Upper Manhattan and higher frequent mental distress coefficients in Lower Manhattan. Conclusion The level of depression incidence does spatially track with predictive health and economic parameters across Manhattan Island. Additional research is encouraged on urban policies that may reduce the mental distress burden on Manhattan residents, as well as investigations of the spatial inversion observed in this study between the exogenous parameters.

Multiscale feature fusion deep network for single image dehazing with continuous memory mechanism

Haze scenes seriously degrade the performance of computer vision systems. Thus, the single image dehazing based on convolutional neural network (CNN), as an important direction of computer vision, has always attracted extraordinary attentions of researchers and made great progress on dehazing effects. However, there are still some non-trivial issues such as uneven dehazing and color distortion, which are not well addressed. To alleviate these issues, this work proposes a new multiscale feature fusion image dehazing network incorporating a contiguous memory mechanism (MFFDN-CM). Specifically, the pixel attention mechanism, continuous memory strategy and residual dense blocks are jointly integrated into the dehazing model with a prevalent encoder-decoder structure(U-Net). Firstly, our model obtains multiscale feature maps by subsampling operations, and further employs skip connections between the corresponding network layers to connect the feature maps between the encoder and the decoder for good feature fusion. Then, we introduce a continuous memory residual block to strengthen the information flows for feature reuse. Moreover, to leverage detail representation and accomplish adaptive dehazing according to the haze density, MFFDN-CM adopts a pixel attention module on the skip connections to combine the residual dense block module of the corresponding decoding layers. Finally, multiple residual blocks are exploited on the bottleneck in encoder-decoder structure to prevent network performance degradation due to vanishing gradients. Experimental results demonstrate the proposed model can achieve better dehazing performance than the state-of-the-art methods based on other deep neural networks.

DCT based densely connected convolutional GRU for real-time speech enhancement

Over the past ten years, deep learning has enabled significant advancements in the improvement of noisy speech. Due to the short time stability of speech signal, previous speech enhancement (SE) methods concentrated only on magnitude estimation, and these methods added a phase of the mixture in reconstructing the speech. The performance is limited in these approaches since the phase will also carry some of the speech information. Some of the speech enhancement approaches were developed later to jointly estimate both magnitudes as well as phases. Recently, complex-valued models, like deep complex convolution recurrent network (DCCRN), are proposed, but the computation of the model is very huge. In this work, we propose a Discrete Cosine Transform-based Densely Connected Convolutional Gated Recurrent Unit (DCTDCCGRU) model using dilated dense block and stacked GRU. The dense connectivity strengthens the gradient propagation by concatenating features from previous layers at the input. The advantage of the dense block is that at various resolutions, the dilated convolutions aid with context aggregation, and the dense connectivity provides a feature map with more precise target information by passing through multiple layers. To represent the correlation between neighboring noisy speech frames, a two Layer GRU is added in the bottleneck of U-Net. The experimental findings demonstrate that the proposed model outperformed the other existing models in terms of STOI (short-time objective intelligibility), PESQ (perceptual evaluation of the speech quality), and output SNR (signal-to-noise ratio).

Automatic detection of aortic valve pathology using 3-chamber cine MRI

Abstract Funding Acknowledgements Type of funding sources: None. Introduction The assessment of aortic valve pathology using magnetic resonance imaging (MRI) typically relies on blood velocity estimates acquired using phase contrast (PC) MRI. However, abnormalities in blood flow through the aortic valve often manifest by the dephasing of blood signal in gated balanced steady-state free precession (bSSFP) scans (cine MRI) [1]. This suggests cine MRI may be used to identify patients with blood flow abnormalities who would benefit from additional detailed PC-MRI scans. Nevertheless, there are currently no tools to automatically identify abnormalities in blood flow through the aortic root in cine MRI which can be used to improve diagnosis and optimise clinical CMR protocols. Aim To determine whether a classification neural network (NN) can automatically detect aortic valve pathology from 3-chamber cine MRI, without the need for dedicated additional imaging or manual annotation. Data We train and test our approach on a retrospective clinical dataset from three UK hospitals, using single-slice bSSFP 3-chamber cine MRI (cardiac frames: 31 ± 15, spatial resolution: 1.17–1.56 x 1.17–1.56 x 8 mm3) from n = 577 patients. Using information from clinical assessments, each image was labelled as: aortic regurgitation (AR) (13%), aortic stenosis (AS) (11%), mixed valve disease (MVD) (9%), or showing no relevant pathology (67%). 322 randomly selected images were used for training, 81 for validation and 174 for testing. Methods We trained a 3D classification NN (DenseNet201 model [2] with 4 dense blocks which include 3D (i.e., spatiotemporal) convolutional layers; Adam optimisation and a cross-entropy loss function) to classify the time-resolved cine MR images. We used binary labels: "pathology" (corresponding to AR, AS or MVD) or "no pathology". As a pre-processing step, we automatically identified the heart in all cine MRI frames using an in-house algorithm that detected motion in the image across the cardiac cycle, and cropped the images to remove non-cardiac anatomical structures. After the classification, we used gradient-weighted class activation mapping (Grad-CAM) [3] to gain confidence that the classification was most dependent on voxels close to the aortic root. Fig 1 shows an overview of the image processing pipeline. Results Our classification model accurately predicts aortic valve pathology using 3-chamber cine MRI: Fig 2 shows the confusion matrix and the evaluation metrics of the study. The Grad-CAM heatmaps showed the aortic region contributed the most to the classification, as seen in Fig 1. Conclusion Convolutional NNs can identify aortic valve pathology automatically from 3-chamber cine MRI, without the need for manual annotations. This can be used to improve diagnosis and optimise clinical CMR protocols, namely as a triage tool to identify the patients most likely to benefit from additional PC-MRI sequences.

FastICENet: A real-time and accurate semantic segmentation model for aerial remote sensing river ice image

River ice semantic segmentation is a crucial task, which can provide us with information for river monitoring, disaster forecasting, and transportation management. Previous works mainly focus on higher accuracy acquirement, while efficiency is also important for reality usage. In this paper, a real-time and accurate river ice semantic segmentation network is proposed, named FastICENet. The general architecture consists of two branches, i.e., a shallow high-resolution spatial branch and a deep context semantic branch, which are carefully designed for the scale diversity and irregular shape of river ice in remote sensing images. Then, a novel Downsampling module and a dense connection block based on a lightweight Ghost module are adopted in the context branch to reduce the computation cost. Furthermore, a learnable upsampling strategy DUpsampling is utilized to replace the commonly used bilinear interpolation to improve the segmentation accuracy. We deploy detailed experiments on three publicly available datasets, named NWPU_YRCC_EX, NWPU_YRCC2, and Alberta River Ice Segmentation Dataset. The experimental results demonstrate that our method achieves state-of-the-art performance with competing methods, on the NWPU_YRCC_EX dataset, we can achieve the segmentation speed as 90.84FPS and the segmentation accuracy as 90.770% mIoU, which also illustrates the good leverage between accuracy and speed. Our code is available at https://github.com/nwpulab113/FastICENet

Denoising distributed acoustic sensing data using unsupervised deep learning

Distributed acoustic sensing (DAS) technology has been widely used in seismic exploration to acquire high-quality data due to its noteworthy advantages, such as high coverage, high resolution, low cost, and strong environmental friendliness. However, the seismic signals acquired in DAS are often masked by various types of noise (e.g., high-frequency random, high-amplitude erratic, horizontal, and coupled noise), which seriously decreases the signal-to-noise ratio. We develop a fully connected neural network with dense and residual connections to attenuate various complex noises in real DAS data. The network is designed to learn the features of useful reflection signals and remove various noises in an unsupervised way, therefore enjoying the convenience of label-free processing. Our network uses several encoders and decoders to compress and reconstruct the abstract waveform features, respectively. Each encoder/decoder consists of one dense block with stacked fully connected blocks (FCBs). To transfer the shallow-level features to the deep level for reuse, we add the skip connections with one FCB between the corresponding encoders and decoders. Our method provides encouraging results when applied to synthetic and real DAS data sets. Compared with several traditional and advanced deep-learning methods, our method can more effectively attenuate strong noise and better extract hidden signals.

SegR-Net: A deep learning framework with multi-scale feature fusion for robust retinal vessel segmentation

Retinal vessel segmentation is an important task in medical image analysis and has a variety of applications in the diagnosis and treatment of retinal diseases. In this paper, we propose SegR-Net, a deep learning framework for robust retinal vessel segmentation. SegR-Net utilizes a combination of feature extraction and embedding, deep feature magnification, feature precision and interference, and dense multiscale feature fusion to generate accurate segmentation masks. The model consists of an encoder module that extracts high-level features from the input images and a decoder module that reconstructs the segmentation masks by combining features from the encoder module. The encoder module consists of a feature extraction and embedding block that enhances by dense multiscale feature fusion, followed by a deep feature magnification block that magnifies the retinal vessels. To further improve the quality of the extracted features, we use a group of two convolutional layers after each DFM block. In the decoder module, we utilize a feature precision and interference block and a dense multiscale feature fusion block (DMFF) to combine features from the encoder module and reconstruct the segmentation mask. We also incorporate data augmentation and pre-processing techniques to improve the generalization of the trained model. Experimental results on three fundus image publicly available datasets (CHASE_DB1, STARE, and DRIVE) demonstrate that SegR-Net outperforms state-of-the-art models in terms of accuracy, sensitivity, specificity, and F1 score. The proposed framework can provide more accurate and more efficient segmentation of retinal blood vessels in comparison to the state-of-the-art techniques, which is essential for clinical decision-making and diagnosis of various eye diseases.

A multi-view CNN encoding for motor imagery EEG signals

Objective:The convolution neural network (CNN) has gained lots of attentions recently in decoding electroencephalogram (EEG) signals in Motor Imagery (MI) Brain–Computer Interface (BCI) designed for improving stroke rehabilitation strategies. However, the extremely non-linear, nonstationary nature of the EEG signals and diversity among individual subjects results in the overfitting of a CNN model and limits its learning ability. In this study, a densely connected convolutional network with multi-view inputs is proposed. Methods:First, different data subsets from the original EEG signals are created as the CNN model inputs through bandpass filters applied to the EEG signals to generate multiple frequency sub-band signals based on brain rhythms. Then, temporal and spatial features are captured based on the whole frequency band and the filtered sub-band signals, respectively. Further, two dense blocks with multi-CNN layers, which connect each layer to every other layer in the feed-forward path, are used to enhance the model learning capabilities and strengthen information propagation. Finally, a concatenation fusion method is used to integrate the extracted features and a fully connected layer for finalizing the classification. Results:The proposed method achieves an average accuracy of 75.16% on the public Korea University EEG dataset which consists the EEG signals of 54 healthy subjects for the two-class motor imagery tasks, higher than other state-of-the-art deep learning methods. Conclusion:The proposed method effectively extracts much richer motor imagery information from the EEG signals in the BCI system and improves the classification accuracy.

Low-Volume Targeted Interscalene Brachial Plexus Block in the Emergency Department as a Safer Alternative for Pain Control for Glenohumeral Reduction: A Case Series

BackgroundAcute glenohumeral dislocation is a common emergency department (ED) presentation, however, pain control to facilitate reduction in these patients can be challenging. Although both procedural sedation and peripheral nerve blocks can provide effective analgesia, both also carry risks. Specifically, the interscalene brachial plexus block carries risk of ipsilateral hemidiaphragmatic paralysis due to inadvertent phrenic nerve involvement. There are techniques, however, that the emergency clinician can utilize to reduce these risks and optimize the interscalene brachial plexus block for specific pathologies such as glenohumeral dislocation. Case SeriesWe report three cases of patients who presented to the ED with acute anterior glenohumeral dislocation. Two of the patients had a history of pulmonary disease. In all three cases, targeted low-volume interscalene nerve blocks were performed and combined with systemic analgesia to facilitate successful closed glenohumeral reduction and reduce the risk of diaphragm paralysis. All 3 patients were monitored after the procedure and discharged from the ED. Why Should an Emergency Physician Be Aware of This?Contrary to anesthesiologists who often seek to obtain dense surgical blocks, the goal of the emergency clinician should be to tailor blocks for specific procedures, patients, and pathologies. The emergency clinician can optimize the interscalene brachial plexus block for glenohumeral dislocation by using a low volume (5–10 mL) of anesthetic targeted to specific nerve roots (C5 and C6) to provide effective analgesia and reduce the risk diaphragm involvement.

Dynamic scene deblurring with continuous cross-layer attention transmission

The deep convolutional neural networks (CNNs) using attention mechanism have achieved great success for dynamic scene deblurring. In most of these networks, only the features refined by the attention maps can be passed to the next layer and the attention maps of different layers are separated from each other, which does not make full use of the attention information from different layers in the CNN. To address this problem, we introduce a new continuous cross-layer attention transmission (CCLAT) mechanism that can exploit hierarchical attention information from all the convolutional layers. Based on the CCLAT mechanism, we use a very simple attention module to construct a novel residual dense attention fusion block (RDAFB). In RDAFB, the attention maps inferred from the outputs of the preceding RDAFB and each layer are directly connected to the subsequent ones, leading to a CCLAT mechanism. Taking RDAFB as the building block, we design an effective architecture for dynamic scene deblurring named RDAFNet. The experiments on benchmark datasets show that the proposed model outperforms the state-of-the-art deblurring approaches, and demonstrate the effectiveness of CCLAT mechanism. The source code is available on: https://github.com/xjmz6/RDAFNet.

Haze removal using deep convolutional neural network for Korea Multi-Purpose Satellite-3A (KOMPSAT-3A) multispectral remote sensing imagery

This paper presents a convolutional neural network to automatically remove the haze distribution using a single multispectral remote sensing image in the raw file format. To train the proposed dehazing network, we synthesized multispectral hazy images using the haze thickness map (HTM) and relative scattering model representing the wavelength-dependent scattering property of the haze distribution. Since the raw multispectral hazy images have a low dynamic range, we cannot accurately estimate the haze distribution directly from them. To differently impose a proper amount of attention to hazy and haze-free regions, we used the HTM from the contrast-enhanced version of the input hazy image. The proposed dehazing network consists of four sub-networks: (i) shallow feature extraction network (SFEN), (ii) cascaded residual dense block network (CRDBN), (iii) multiscale feature extraction network (MFEN), and (iv) refinement network (RN). The densely connected convolutional layers and local residual learning allow the residual dense block (RDB) to extract the abundant local features, and the cascaded architecture further improves the propagation of the local information and gradients. The MFEN is used to extract multiscale local features representing the hierarchical information for the haze distribution and haze-free region. Experimental results demonstrated that the proposed method can achieve improved dehazing performance on Korea Multi-Purpose Satellite-3A (KOMPSAT-3A) multispectral remote sensing imagery without undesired artifacts. In the sense of quantitative assessment, the proposed method produced improved peak signal-to-noise ratio (PSNR) by 10%, structural similarity index measure (SSIM) by 1%, and spectral angle mapper (SAM) by 19% compared with the existing best method.

Rate-Distortion Optimization-Based Learned Fractional Interpolation Filter Design for HEVC

Fractional motion compensation (FMC) has been widely adopted in modern video coding standards such as high-efficiency video coding (HEVC). The fractional pixels in FMC are generated by the interpolation filters, and the coefficients of the filters are designed using a Fourier decomposition of the discrete cosine transform (DCT) in HEVC. To overcome the compression deficiency caused by these handcrafted filters, deep learning-based interpolation filter approaches have been introduced recently. In this paper, we analyze the drawbacks of the mean square error (MSE) only loss function which is widely adopted in existing learned-based approaches, and introduce the rate-distortion optimization (RDO) theory into network training to formulate a joint loss function to overcome this issue. Then, a new training scheme is proposed for the detailed implementation of the joint loss function. Besides the calculated MSE loss, the residual will be sent into a rate estimation model which includes three stages: transform/quantization/rate modeling, and the output rate cost will be used as the second part of the loss function. Finally, we propose a novel network structure that utilizes the deformable convolution layer and the residual dense block to enhance the feature extraction and transmission ability to adapt to complex video contents. The proposed RDO-based interpolation filter (RDOIF) has been integrated into HEVC reference software HM-16.7, and extensive experiments are conducted to verify the efficiency of the proposed method. Experimental results show that the proposed RDOIF method achieves 3.1%, 2.4%, and 1.4% BD-rate reduction on average under the low-delay-P, low-delay-B, and random access configurations, respectively. When implemented in the versatile video coding (VVC) test model VTM-10.0, the proposed method can achieve 2.3%, 1.2%, and 0.5% BD-rate savings on average under the low-delay-P, low-delay-B, and random access configurations, respectively.

Single‐image super‐resolution using lightweight transformer‐convolutional neural network hybrid model

AbstractWith constant advances in deep learning methods as applied to image processing, deep convolutional neural networks (CNNs) have been widely explored in single‐image super‐resolution (SISR) problems and have attained significant success. These CNN‐based methods cannot fully use the internal and external information of the image. The authors add a lightweight Transformer structure to capture this information. Specifically, the authors apply a dense block structure and residual connection to build a residual dense convolution block (RDCB) that reduces the parameters somewhat and extracts shallow features. The lightweight transformer block (LTB) further extracts features and learns the texture details between the patches through the self‐attention mechanism. The LTB comprises an efficient multi‐head transformer (EMT) with small graphics processing unit (GPU) memory footprint, and benefits from feature preprocessing by multi‐head attention (MA), reduction, and expansion. The EMT significantly reduces the use of GPU resources. In addition, a detail‐purifying attention block (DAB) is proposed to explore the context information in the high‐resolution (HR) space to recover more details. Extensive evaluations of four benchmark datasets demonstrate the effectiveness of the authors’ proposed model in terms of quantitative metrics and visual effects. The proposed EMT only uses about 40% as much GPU memory as other methods, with better performance.

A Single Image Dehazing Method Based on End-to-End CPAD-Net Network in Deep Learning Environment

To address the issues of blurred details and distortion of color in the images recovered by the original AOD-Net dehazing method, this paper proposes a CPAD-Net dehazing network model based on attention mechanism and dense residual blocks. The network is improved on the basis of AOD-Net, which can reduce the errors arising from the separately determined transmittance and atmospheric light values. A new dense residual block structure is designed to replace the traditional convolution method, which effectively improves the detail processing capability and the representation ability of the network model for image feature information. On this basis, the attention module determines how to learn the weights according to the feature importance of distinct channels and distinct pixels, and then obtain the recovery of images in terms of color and texture. The experiments showed that the dehazing efficiency of our method are richer in texture detail information and more natural in color recovery. Compared with other algorithms, the PSNR and SSIM indexes of our method are considerably superior to those listed algorithms, which definitively demonstrates that the dehazing effect of our method is more effective, and the recovered images are more realistic and natural.

Underwater Image Enhancement via Triple-Branch Dense Block and Generative Adversarial Network

The complex underwater environment and light scattering effect lead to severe degradation problems in underwater images, such as color distortion, noise interference, and loss of details. However, the degradation problems of underwater images bring a significant challenge to underwater applications. To address the color distortion, noise interference, and loss of detail problems in underwater images, we propose a triple-branch dense block-based generative adversarial network (TDGAN) for the quality enhancement of underwater images. A residual triple-branch dense block is designed in the generator, which improves performance and feature extraction efficiency and retains more image details. A dual-branch discriminator network is also developed, which helps to capture more high-frequency information and guides the generator to use more global content and detailed features. Experimental results show that TDGAN is more competitive than many advanced methods from the perspective of visual perception and quantitative metrics. Many application tests illustrate that TDGAN can significantly improve the accuracy of underwater target detection, and it is also applicable in image segmentation and saliency detection.

RaGMAN: A relativistic average generative multi-adversarial network for pansharpening

Pansharpening aims to reconstruct a high-resolution multi-spectral image (HRMS) from a low spatial resolution multi-spectral (MS) image and a high spatial resolution single band panchromatic (PAN) image acquired from the same satellite. To solve the spatial information and spectral information of the fused image that are not evenly preserved, this paper proposes a relativistic average generative multi-adversarial network (RaGMAN) for pansharpening. The RaGMAN consists of two parts. One builds a two-stream generator network to generate HRMS images by extracting features from PAN and MS images. The other is a dual discriminator to preserve the spectral and spatial information of the input when performing fusion. To make the generated image contain more spatial information, we propose multiple residual dense block in the generator. At the same time, in order to improve the overall quality of the fused images, two relativistic average discriminators are used in the network. Furthermore, a novel hybrid loss function is introduced to optimize training. Compared with seven state-of-the-art methods on GF-2, QB and WV-3 datasets, experimental results show that the proposed RaGMAN method can produce excellent pansharpening performance in terms of spatial and spectral fidelity.

STFRDN: a residual dense network for remote sensing image spatiotemporal fusion

ABSTRACT Convolutional Neural Networks (CNN) are useful models for spatiotemporal fusion, especially under strong temporal changes. Convolutional layers with receptive fields of various sizes produce hierarchical features that can improve prediction performance. In this study, we aimed to develop a CNN model for spatiotemporal fusion, which can extract hierarchical features and fully use them. The proposed network is composed of residual dense blocks with local dense connections that effectively utilize the hierarchical features. We conducted the experiments in the newly proposed Kansas dataset, which includes Sentinel-2 and Sentinel-3 image pairs with large resolution differences and strong temporal changes. Based on both quantitative and qualitative evaluation, experiments revealed that the STFRDN model outperformed the Flexible Spatiotemporal DAta Fusion (FSDAF) 2.0, Reliable and Adaptive Spatiotemporal Data Fusion (RASDF), Fit-FC, and DMNet methods in all of the test groups. The second major finding was that the STFRDN model was able to produce sufficiently successful predictions against cases with such large resolution differences. To get more accurate prediction results when there is a significant resolution difference, it is important to extract the right features that can model the discrepancy. Loss of spatial detail can be avoided by making full use of hierarchical features without any manipulation after extraction. The results of this study provide insights into the significance of extracting and fully utilizing the hierarchical features.

An adaptive weighted attention-enhanced deep convolutional neural network for classification of MRI images of Parkinson's disease

BackgroundParkinson's disease (PD) is the second prevalent neurological diseases with a significant growth rate in incidence. Convolutional neural networks using structural magnetic resonance images (sMRI) are widely used for PD classification. However, the areas of change in the patient's MRI images are small and unfixed. Thus, capturing the features of the areas accurately where the lesions changed became a problem. MethodWe propose a deep learning framework that combines multi-scale attention guidance and multi-branch feature processing modules to diagnose PD by learning sMRI T2 slice features. In this scheme, firstly, to achieve effective feature transfer and gradient descent, a deep convolutional neural network framework based on dense block is designed. Next, an Adaptive Weighted Attention algorithm is proposed, whose pursers is to extract multi branch and even diverse features. Finally, Dropout layer and SoftMax layer are added to the network structure to obtain good classification results and rich and diverse feature information. The Dropout layer is used to reduce the number of intermediate features to increase the orthogonality between features of each layer. The activation function SoftMax increases the flexibility of the neural network by increasing the degree of fitting to the training set and converting linear to nonlinear. ResultsThe best performance of the proposed method an accuracy of 92%, a sensitivity of 94%, specificity of 90% and a F1 score of 95% respectively for identifying PD and HC. ConclusionExperiments show that the proposed method can successfully distinguish PD and NC. Good classification results were obtained in PD diagnosis classification task and compared with advanced research methods.

Multi-classification deep neural networks for identification of fish species using camera captured images.

Regular monitoring of the number of various fish species in a variety of habitats is essential for marine conservation efforts and marine biology research. To address the shortcomings of existing manual underwater video fish sampling methods, a plethora of computer-based techniques are proposed. However, there is no perfect approach for the automated identification and categorizing of fish species. This is primarily due to the difficulties inherent in capturing underwater videos, such as ambient changes in luminance, fish camouflage, dynamic environments, watercolor, poor resolution, shape variation of moving fish, and tiny differences between certain fish species. This study has proposed a novel Fish Detection Network (FD_Net) for the detection of nine different types of fish species using a camera-captured image that is based on the improved YOLOv7 algorithm by exchanging Darknet53 for MobileNetv3 and depthwise separable convolution for 3 x 3 filter size in the augmented feature extraction network bottleneck attention module (BNAM). The mean average precision (mAP) is 14.29% higher than it was in the initial version of YOLOv7. The network that is utilized in the method for the extraction of features is an improved version of DenseNet-169, and the loss function is an Arcface Loss. Widening the receptive field and improving the capability of feature extraction are achieved by incorporating dilated convolution into the dense block, removing the max-pooling layer from the trunk, and incorporating the BNAM into the dense block of the DenseNet-169 neural network. The results of several experiments comparisons and ablation experiments demonstrate that our proposed FD_Net has a higher detection mAP than YOLOv3, YOLOv3-TL, YOLOv3-BL, YOLOv4, YOLOv5, Faster-RCNN, and the most recent YOLOv7 model, and is more accurate for target fish species detection tasks in complex environments.