Search

Abstract Magnetic resonance imaging is a medical imaging technique to create comprehensive images of the tissues and organs in the body. This study presents an advanced approach for storing and compressing neuroimaging informatics technology initiative files, a standard format in magnetic resonance imaging. It is designed to enhance telemedicine services by facilitating efficient and high-quality communication between healthcare practitioners and patients. The proposed downsampling approach begins by opening the neuroimaging informatics technology initiative file as volumetric data and then planning it into several slice images. Then, the quantization hiding technique will be applied to each of the two consecutive slice images to generate the stego slice with the same size. This involves the following major steps: normalization, microblock generation, and discrete cosine transformation. Finally, it assembles the resultant stego slice images to produce the final neuroimaging informatics technology initiative file as volumetric data. The upsampling process, designed to be completely blind, reverses the downsampling steps to reconstruct the subsequent image slice accurately. The efficacy of the proposed method was evaluated using a magnetic resonance imaging dataset, focusing on peak signal-to-noise ratio, signal-to-noise ratio, structural similarity index, and Entropy as key performance metrics. The results demonstrate that the proposed approach not only significantly reduces file sizes but also maintains high image quality.

Crowd counting is an important research topic in the fields of computer vision and image processing, with monitoring and management of crowded scenes becoming an increasingly prominent issue. Existing methods still suffer from the problem of severe overlap in density maps within dense areas, leading to inadequate counting and localization accuracy. This paper presents innovative research on crowd counting and localization. Firstly, addressing the limitations of density maps in localization performance in existing algorithms, we optimize the generation method of FIDT maps, decoupling the counting and localization tasks. By avoiding the problem of overlap in dense areas, the optimized label maps achieve a good balance between counting accuracy and localization, with MAE and MSE reaching 64.1 and 103.9 in SHHA, and 10.9 and 17.4 in SHHB, respectively.Secondly, to address the scale insensitivity of the encoder and the potential loss of critical features during the encoding process, we propose the Adaptive Feature Fusion Module and the Multi-Scale Global Attention Upsampling Module, constructing the CALNET network. By reducing redundant features inside and outside the separable branch, the model achieves global fusion of shallow features during the decoding process. The F1-m scores obtained on the SHHA and SHHB datasets reach 72.9% and 79.4% respectively, significantly improving the model’s performance.Finally, this paper extends the application of crowd counting and localization algorithms to different domains such as citrus orchards, vehicles, and campus crowds. Through experiments, the robustness and transferability of the network are validated, expanding the application areas of crowd counting and localization algorithms and providing a broader space for future research.

The current advanced neural network models are expanding in size and complexity to achieve improved detection accuracy. This study designs a lightweight fabric defect detection algorithm based on YOLOv7-tiny, called YOLOv7-tiny-MGCK. Its objectives are to improve the performance of fabric defect detection against complex backgrounds and to find a balance between the algorithm’s lightweight nature and its accuracy. The algorithm utilizes the Mish activation function, known for its superior nonlinear performance capability and smoother curve, enabling the neural network to manage more complex challenges. The Ghost convolution module is also incorporated to reduce computation and model parameters. The lightweight upsampling technique CARAFE facilitates the flexible extraction of deep features, coupled with their integration with shallow features. In addition, an improved K-Means clustering algorithm, KMMP, is employed to select appropriate anchor box for fabric defects. The experimental results show: a reduction in the number of parameters by 45.5% and computational volume by 41.0%, along with increases in precision by 3.9%, recall by 7.0%, and mAP by 3.0%. These results indicated that the improved algorithm achieves a more effective balance between detection performance and the requirement for a lightweight solution.

Point cloud super-resolution plays a central role in the mesh's quality in 3D reconstruction, while the feature extractor is vital for the learning-based point cloud upsampling pipelines. In this paper, we propose an arbitrary 3D point cloud upsampling network (PSR-GAT), which comprises the feature extraction module, GAT module, and upsampling module. For the input point cloud, the feature extraction module locates k nearest points of each point in 3D space by k-NN algorithm, then converts the local geometry information into high dimensional feature space through a multi-layer point-wise convolution. The GAT module converts the local geometry feature of each point into the semantic feature through a multi-layer graph attention network. The module dynamically adjusts the neighbor space of the point in each layer to increase the receptive field range and effectively fuses the semantic information of different levels through residual connection. This makes the local geometric in- formation extraction efficient. The upsampling module adds the number of points and maps them from feature space to 3D space. Extensive experimental results show that PSR-GAT exhibits a better performance than the existing state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Magnetic resonance imaging is a medical imaging technique to create comprehensive images of the tissues and organs in the body. This study presents an advanced approach for storing and compressing neuroimaging informatics technology initiative files, a standard format in magnetic resonance imaging. It is designed to enhance telemedicine services by facilitating efficient and high-quality communication between healthcare practitioners and patients. The proposed downsampling approach begins by opening the neuroimaging informatics technology initiative file as volumetric data and then planning it into several slice images. Then, the quantization hiding technique will be applied to each of the two consecutive slice images to generate the stego slice with the same size. This involves the following major steps: normalization, microblock generation, and discrete cosine transformation. Finally, it assembles the resultant stego slice images to produce the final neuroimaging informatics technology initiative file as volumetric data. The upsampling process, designed to be completely blind, reverses the downsampling steps to reconstruct the subsequent image slice accurately. The efficacy of the proposed method was evaluated using a magnetic resonance imaging dataset, focusing on peak signal-to-noise ratio, signal-to-noise ratio, structural similarity index, and Entropy as key performance metrics. The results demonstrate that the proposed approach not only significantly reduces file sizes but also maintains high image quality. [ABSTRACT FROM AUTHOR]

Mobile LiDAR technology is a powerful tool that accurately captures spatial information about typical static objects in road scenes. However, the precise extraction and classification of these objects pose persistent technical challenges. In this paper, we employ a deep learning approach to tackle the point cloud classification problem. Despite the popularity of the PointNet++ network for direct point cloud processing, it encounters issues related to insufficient feature learning and low accuracy. To address these limitations, we introduce a novel layer-wise optimization network, LO-Net. Initially, LO-Net utilizes the set abstraction module from PointNet++ to extract initial local features. It further enhances these features through the edge convolution capabilities of GraphConv and optimizes them using the "Unite_module" for semantic enhancement. Finally, it employs a point cloud spatial pyramid joint pooling module, developed by the authors, for the multiscale pooling of final low-level local features. Combining three layers of local features, LO-Net sends them to the fully connected layer for accurate point cloud classification. Considering real-world scenarios, road scene data often consist of incomplete point cloud data due to factors such as occlusion. In contrast, models in public datasets are typically more complete but may not accurately reflect real-world conditions. To bridge this gap, we transformed road point cloud data collected by mobile LiDAR into a dataset suitable for network training. This dataset encompasses nine common road scene features; hence, we named it the Road9 dataset and conducted classification research based on this dataset. The experimental analysis demonstrates that the proposed algorithm model yielded favorable results on the public datasets ModelNet40, ModelNet10, and the Sydney Urban Objects Dataset, achieving accuracies of 91.2%, 94.2%, and 79.5%, respectively. On the custom road scene dataset, Road9, the algorithm model proposed in this paper demonstrated outstanding classification performance, achieving a classification accuracy of 98.5%. [ABSTRACT FROM AUTHOR]

Pan-sharpening is a procedure to fuse the spatial detail of high-resolution multispectral images (HR-MSI) and low-resolution hyperspectral images (LR-HSI) to produce HR-MSI. Due to increase in high-resolution satellites, methods based on pan-sharpening are increasingly utilized all over the world. However, the majority of techniques consider pan-sharpening as a major issue, which hinders the discriminative ability. This work proposes an optimization-based deep model for pan-sharpening using LR-HSI and HR-MSI. Initially, the LR-HSI is input into an up-sampling mode, and the resulting image is fed into weighted linear regression. Concurrently, HR-MSI is supplied into weighted linear regression. Weighted linear regression is used to combine the upsampled LR-HSI and HR-MSI. The HR-MSI is then sent into the Deep Maxout network (DMN), which learns the priors via residual learning. Furthermore, the suggested Competitive Multi-Verse Feedback Artificial Tree (CMVFTA) strategy is used for DMN training, which is constructed by combining the Competitive Multi-Verse Optimizer (CMVO) and Feedback Artificial Tree (FAT) approaches. Finally, the DMN, LR-HSI, and HR-MSI outputs are merged together to provide a pan-sharpening image. The proposed CMVFTA-based DMN offered enhanced performance with Degree of Distortion (DD) of 0.0402 dB, Peak signal-to-noise ratio (PSNR) of 49.60 dB, Root Mean Squared Error (RMSE) of 0.330, Relative Average Spectral Error (RASE) of 0.322, Filtered Correlation Coefficients (FCC) of 0.874, Quality with no reference (QNR) of 76.19. [ABSTRACT FROM AUTHOR]

High resolution remains a primary goal in the advancement of synthetic aperture radar (SAR) technology. The backprojection (BP) algorithm, which does not introduce any approximation throughout the imaging process, is broadly applicable and effectively meets the demands for high-resolution imaging. Nonetheless, the BP algorithm necessitates substantial interpolation during point-by-point processing, and the precision and effectiveness of current interpolation methods limit the imaging performance of the BP algorithm. This paper proposes a TSU-ICSI (Time-shift Upsampling-Improved Cubic Spline Interpolation) interpolation method that integrates time-shift upsampling with improved cubic spline interpolation. This method is applied to the BP algorithm and presents an efficient implementation method in conjunction with the GPU architecture. TSU-ICSI not only maintains the accuracy of BP imaging processing but also significantly boosts performance. The effectiveness of the BP algorithm based on TSU-ICSI is confirmed through simulation experiments and by processing measured data collected from both airborne SAR and spaceborne SAR. [ABSTRACT FROM AUTHOR]

This paper introduces a novel edge detection operator called 'vision transformer-based Dexinet (ViT-DexiNet)' to address the challenges of detecting small objects in synthetic aperture radar (SAR) images. SAR images are typically impacted by strong multiplicative noise, making edge detection difficult. Existing traditional methods have limited spectral data preservation capabilities and often result in a loss of clarity and integrity of salient features in SAR images. The proposed ViT-DexiNet operator employs a series of interconnected layers to extract and refine salient edge features from SAR images. It utilizes a vision transformer self-attention layer to capture the pattern and structural details of image features crucial for determining edges. The extracted feature maps are then processed by the DexiNet architecture, which consists of a dense block, transfer block, and upsampling network. This architecture helps preserve edge information at different scales in deeper layers. The series of layered blocks generate edge maps, which are concatenated and averaged through smoothing to remove noise and enhance edge details in SAR images, resulting in a final high-quality edge map. To evaluate the proposed ViT-DexiNet method, both qualitative and quantitative analyses are conducted using standard edge detection operators such as Canny and Sobel. The empirical results demonstrate that the ViT-DexiNet surpasses baseline edge detection operators. The achieved values of proposed edge detection operator are 97.92%, 97.72%, 97.64% and 97.41%, respectively for the metrics accuracy, precision, recall and f1-score. The ViT-DexiNet offers high-quality edge maps, simplifying the interpretation of data for small object detection. Overall, the ViT-DexiNet method shows promise in overcoming the limitations of traditional approaches and improving the detection of edges in SAR images. [ABSTRACT FROM AUTHOR]

Point cloud upsampling can improve the resolutions of point clouds and maintain the forms of point clouds, which has attracted more and more attention in recent years. However, upsampling networks sometimes generate point clouds with unclear contours and deficient topological structures, i.e., the problem of insufficient form fidelity of upsampled point clouds. This paper focuses on the above problem. Firstly, we manage to find the points located at contours or sparse positions of point clouds, i.e., the form describers, and make them multiply correctly. To this end, 3 statistics of points, i.e., local coordinate difference, local normal difference and describing index, are designed to estimate the form describers of the point clouds and rectify the feature aggregation of them with reliable neighboring features. Secondly, we divide points into disjoint levels according to the above statistics and apply K nearest neighbors algorithm to the points of different levels respectively to build an accurate graph. Finally, cascaded networks and graph information are fused and added to the feature aggregation so that the network can learn the topology of objects deeply, enhancing the perception of model toward graph information. Our upsampling model PU-FPG is obtained by combining these 3 parts with upsampling networks. We conduct abundant experiments on PU1K dataset and Semantic3D dataset, comparing the upsampling effects of PU-FPG and previous works in multiple metrics. Compared with the baseline model, the Chamfer distance, the Hausdorff distance and the point-to-surface distance of PU-FPG are reduced by 0.159 × 10-3, 2.892 × 10-3 and 0.852 × 10-3, respectively. This shows that PU-FPG can improve the form fidelity and raise the quality of upsampled point clouds effectively. Our code is publicly available at . [ABSTRACT FROM AUTHOR]

Nowadays, deep convolutional neural networks (CNNs) are mostly applied for image Super-Resolution (SR). But still, there are some disadvantages of using CNN as it causes enormous computational complexity as if it is directly applied to SR applications. In this paper, dilated convolution is adopted that expands receptive field without any pixel information losses. The dilated convolution is designed as recursive residual network; therefore, internal parameters are preserved. Therefore, the model is termed as Multilevel Progressive Recursive Dilated Networks with Correlation Filter (MPRDNCF) and adopted progressive approach with different levels of recursive dilated residual network that is interleaved with correlation filter for upscaling of image. This module upscales with different scaling factors and magnifies it using deconvolution layer. MPRDNCF model used progressive recursive dilated residual learning approach which shares the information between the convolution layers for the identity prior during the network training. The architecture of MPRDNCF is 33 layers of CNN. We have presented an ablation study on Set5, Set14, Urban100, and BSD100 datasets, and also presented its superior result with comparison to the existing technique of state-of-art. [ABSTRACT FROM AUTHOR]

The prevalence of heart disease has been increasing in the last ten years. One of the cardiac diagnostic tools is echocardiography. Echocardiogram medical images provide essential information, including shape, size, pumping capacity, heart function abnormalities, and location of heart damage, but echocardiogram images have high noise content and poor contrast, as well as limitations due to differences in anatomy or body mass. This will affect the reading results of patient diagnosis. Therefore, image quality improvement is needed by removing noise and increasing image contrast. This research has improved image quality using a method with low mathematical complexity and a fast computational process. The method used is the Upsampling method to generate a reference image. The quality of the image produced was the Nearest Neighbor upsampling method: 2.8 dB, Bi-linear Interpolation: 2.78 dB, and Bi-cubic Interpolation: 2.73 dB. Furthermore, the image with the highest SNR value is processed with Histogram Matching to accelerate improving image quality. The Histogram Matching image increases quality by more than 50% with a SSIM value of 0.54. The required computational process to apply this method to each medical image has an average duration of 0.4 s. This result provides a higher value than several methods using linear scaling and speckle reducing.

In a foggy traffic environment, the vision sensor signal of intelligent vehicles will be distorted, the outline of obstacles will become blurred, and the color information in the traffic road will be missing. To solve this problem, four ultra-fast defogging strategies in a traffic environment are proposed for the first time. Through experiments, it is found that the performance of Fast Defogging Strategy 3 is more suitable for fast defogging in a traffic environment. This strategy reduces the original foggy picture by 256 times via bilinear interpolation, and the defogging is processed via the dark channel prior algorithm. Then, the image after fog removal is processed via 4-time upsampling and Gaussian transform. Compared with the original dark channel prior algorithm, the image edge is clearer, and the color information is enhanced. The fast defogging strategy and the original dark channel prior algorithm can reduce the defogging time by 83.93–84.92%. Then, the image after fog removal is inputted into the YOLOv4, YOLOv5, YOLOv6, and YOLOv7 target detection algorithms for detection and verification. It is proven that the image after fog removal can effectively detect vehicles and pedestrians in a complex traffic environment. The experimental results show that the fast defogging strategy is suitable for fast defogging in a traffic environment. [ABSTRACT FROM AUTHOR]

The projection of a point cloud onto a 2D camera image is relevant in the case of various image analysis and enhancement tasks, e.g., (i) in multimodal image processing for data fusion, (ii) in robotic applications and in scene analysis, and (iii) for deep neural networks to generate real datasets with ground truth. The challenges of the current single-shot projection methods, such as simple state-of-the-art projection, conventional, polygon, and deep learning-based upsampling methods or closed source SDK functions of low-cost depth cameras, have been identified. We developed a new way to project point clouds onto a dense, accurate 2D raster image, called Triangle-Mesh-Rasterization-Projection (TMRP). The only gaps that the 2D image still contains with our method are valid gaps that result from the physical limits of the capturing cameras. Dense accuracy is achieved by simultaneously using the 2D neighborhood information (rx , ry) of the 3D coordinates in addition to the points P (X , Y , V) . In this way, a fast triangulation interpolation can be performed. The interpolation weights are determined using sub-triangles. Compared to single-shot methods, our algorithm is able to solve the following challenges. This means that: (1) no false gaps or false neighborhoods are generated, (2) the density is XYZ independent, and (3) ambiguities are eliminated. Our TMRP method is also open source, freely available on GitHub, and can be applied to almost any sensor or modality. We also demonstrate the usefulness of our method with four use cases by using the KITTI-2012 dataset or sensors with different modalities. Our goal is to improve recognition tasks and processing optimization in the perception of transparent objects for robotic manufacturing processes. [ABSTRACT FROM AUTHOR]

In pedestrian detection, the upsampling operation of YOLOv4 during feature aggregation affects the integrity of feature information for small-scale and occluded targets. To address this issue, we propose a pedestrian detection model named Shuffle Expansion YOLOv4 (SE-YOLOv4) composed of a path aggregation network based on PixelShuffle (Shuffle-PANet) and an efficient pyramid atrous convolutional block attention module (EPA-CBAM), to improve the detection performance of small-scale and occluded pedestrian targets. First, we propose a feature aggregation network Shuffle-PANet based on PixelShuffle to maintain the feature information integrity of small-scale and occluded targets by expanding high-resolution feature maps through convolutions and interchannel periodic shuffling instead of linear interpolation-based upsampling. Then, we propose EPA-CBAM, whose channel attention module (EPA-CAM) can build a pyramid structure and obtain fine-grained multiscale spatial information in different channels by dilated convolutions of corresponding sizes. The results show that the miss rate of SE-YOLOv4 decreased by 3.54% compared with that of the vanilla YOLOv4 on the CityPersons dataset. Comparison experiment results on four challenging pedestrian detection datasets show that our method achieves very competitive performance and maintains a reasonable balance between accuracy and speed. [ABSTRACT FROM AUTHOR]

We present a simple but effective technique to smooth out textures while preserving the prominent structures. Our method is built upon a key observation---the coarsest level in a Gaussian pyramid often naturally eliminates textures and summarizes the main image structures. This inspires our central idea for texture filtering, which is to progressively upsample the very low-resolution coarsest Gaussian pyramid level to a full-resolution texture smoothing result with well-preserved structures, under the guidance of each fine-scale Gaussian pyramid level and its associated Laplacian pyramid level. We show that our approach is effective to separate structure from texture of different scales, local contrasts, and forms, without degrading structures or introducing visual artifacts. We also demonstrate the applicability of our method on various applications including detail enhancement, image abstraction, HDR tone mapping, inverse halftoning, and LDR image enhancement. Code is available at https://rewindl.github.io/pyramid_texture_filtering/. [ABSTRACT FROM AUTHOR]

Constrained by current sensing technology, depth camera only acquires a low-resolution depth image that does not meet actual requirements. To solve this problem, this paper take a divide-and-conquer strategy to synthesize a high-resolution depth image from a low-resolution range image under the guidance of a registered high-resolution color image. Initially, the depth image is divided into planar areas and edge regions. For different zones, we exploit different methods to interpolate the missing depths. At planar area, the linear interpolation method is employed to perform upsampling. At edge region, a segmentation-separation upsampling method is used to interpolate the missing values. Then the upsampling result are refined on the Depth CNN that is built in this paper. We conduct extensive experiments on the benchmark database and real world data with various upsampling rates to illustrate the upsampling ability of our method. The comparison with classical super-resolution algorithms demonstrates that our upsampling algorithm achieves the best quality with fewer artifacts and our depth CNN outperforms the most state-of-the-art methods in terms of qualitative and quantitative evaluations.

Mobile LiDAR technology is a powerful tool that accurately captures spatial information about typical static objects in road scenes. However, the precise extraction and classification of these objects pose persistent technical challenges. In this paper, we employ a deep learning approach to tackle the point cloud classification problem. Despite the popularity of the PointNet++ network for direct point cloud processing, it encounters issues related to insufficient feature learning and low accuracy. To address these limitations, we introduce a novel layer-wise optimization network, LO-Net. Initially, LO-Net utilizes the set abstraction module from PointNet++ to extract initial local features. It further enhances these features through the edge convolution capabilities of GraphConv and optimizes them using the “Unite_module” for semantic enhancement. Finally, it employs a point cloud spatial pyramid joint pooling module, developed by the authors, for the multiscale pooling of final low-level local features. Combining three layers of local features, LO-Net sends them to the fully connected layer for accurate point cloud classification. Considering real-world scenarios, road scene data often consist of incomplete point cloud data due to factors such as occlusion. In contrast, models in public datasets are typically more complete but may not accurately reflect real-world conditions. To bridge this gap, we transformed road point cloud data collected by mobile LiDAR into a dataset suitable for network training. This dataset encompasses nine common road scene features; hence, we named it the Road9 dataset and conducted classification research based on this dataset. The experimental analysis demonstrates that the proposed algorithm model yielded favorable results on the public datasets ModelNet40, ModelNet10, and the Sydney Urban Objects Dataset, achieving accuracies of 91.2%, 94.2%, and 79.5%, respectively. On the custom road scene dataset, Road9, the algorithm model proposed in this paper demonstrated outstanding classification performance, achieving a classification accuracy of 98.5%.

This paper proposes a framework that enables the online generation of virtual point clouds relying only on previous camera and point clouds and current camera measurements. The continuous usage of the pipeline generating virtual LIDAR measurements makes the temporal up-sampling of point clouds possible. The only requirement of the system is a camera with a higher frame rate than the LIDAR equipped to the same vehicle, which is usually provided. The pipeline first utilizes optical flow estimations from the available camera frames. Next, optical expansion is used to upgrade it to 3D scene flow. Following that, ground plane fitting is made on the previous LIDAR point cloud. Finally, the estimated scene flow is applied to the previously measured object points to generate the new point cloud. The framework's efficiency is proved as state-of-the-art performance is achieved on the KITTI dataset. [ABSTRACT FROM AUTHOR]

It is common in eye movement acquisition that data comes from different devices and exhibits different sampling rates. One of the solutions to this problem is to recalculate the signal into another sampling rate. When upsampling is performed on a sequence of samples, it approximates the sequence that would have been obtained by sampling the signal at a higher rate. Many methods may be used for eye movement signal upsampling. Recently, Convolutional Neural Networks proved to be very efficient in upsampling (or supersampling) digital images. This paper attempts to utilize Convolutional Neural Networks using two architectures with different types of layers on the signal sampled with 125 Hz to obtain an eight-time increase in the sampling rate and produce the signal with a 1000 Hz sampling rate. The experiments on the GazeBase dataset proved that this solution is feasible, and the Convolutional Neural Network can learn the characteristic of the eye movement signal. [ABSTRACT FROM AUTHOR]

Spatial prediction(SP) based on machine learning(ML) has been applied to soil water quality, air quality, marine environment, etc. However, there are still deficiencies in dealing with the problem of small samples. Normally, ML requires large amounts of training samples to prevent underfitting. And the data augmentation(DA) methods of mixup and synthetic minority over-sampling technique(SMOTE) ignore the similarity of geographic information. Therefore, this paper proposes a modified upsampling method and combines it with the random forest spatial interpolation(RFSI) to deal with the small sample problem in geographical space. The modified upsampling is mainly reflected in the following two aspects. Firstly, in the process of selecting the nearest points, it is to select points with similar geographic information in some aspects of the category after classification. Secondly, the selected difference is the difference of each category. In order to verify the effectiveness of the proposed method, we use daily precipitation data for January 2018 in Chongqing. The experimental results show that the combination of the modified upsampling method and RFSI effectively improves the accuracy of SP. [ABSTRACT FROM AUTHOR]

Potato is one of the most typical shallow root crops. The root distribution characteristics can then dominate the effective water and nutrient management. The accurate segmentation of root system can be the essential prerequisite for the key parameters of root system structure. Taking the potato root images as the research object, this study aims to achieve the non-contact, low-cost, fast, and accurate segmentation of potato root images. A potato root image segmentation was also proposed to monitor the growth state of potato using the improved DeepLabv3+ semantic segmentation network. The root length of the output image was then calculated. The spatial and temporal dynamic distribution characteristics of potato roots were calculated in the northern foothills of Yinshan Mountain in Inner Mongolia, China. The test results show that the training time of the improved MobileNetv2 was only 10.05h, which was 2.27 and 4.1h less than ResNet50, and Xception, respectively. In terms of the image segmentation performance, the MIoU of MobileNetv2 reached 92.26%, which was 1.84 and 2.68 percentage points higher than that of ResNet50 and Xception, respectively. The MPA reached 94.15%, which was 1.78 and 2.69 percentage points higher than that of the ResNet50 and Xception, respectively. The MIoU and MPA of DeepLabv3+ increased by 1.43 and 1.62 percentage points after the introduction of the improved MobileNetv2 backbone network, according to the standard DeepLabv3+. The MIoU and MPA increased by 1.61 and 1.80 percentage points after the introduction of CARAFE upsampling, respectively. The MIoU and MPA were improved by 2.92 and 2.68 percentage points, respectively, after the introduction of the CBAM attention mechanism. Furthermore, the CARAFE upsampling and CBAM attention mechanisms were introduced to improve the MobileNetv2 backbone network for the combination of different modules. After that, the MIoU and MPA increased by 2.30 and 2.61 percentage points, and 3.02 and 2.82 percentage points, respectively. The MIoU and MPA of the CARAFE upsampling increased by 2.98 and 2.23 percentage points, respectively, after combining with the CBAM attention mechanism. The best three improvement strategies were selected to increase the MIoU and MPA by 4.18 and 4.28 percentage points, respectively. The MIoU and MPA of the improved DeepLabv3+ model were 94.05% and 95.72%, respectively. The MIoU increased by 6.67, 4.92, 8.80 and 4.21 percentage points, respectively, compared with the SegNet, PSPNet, U-Net and standard DeepLabv3+, and the MPA increased by 6.7, 4.86, 8.25, and 4.53 percentage points, respectively. The training time was 9.52h, which was shortened by 6.8, 3.99, 4.56, and 3.94h, respectively, compared with the SegNet, PSPNet, U-Net, and standard DeepLabv3+. The FLOPs were reduced by 45×109, 34×109, 29×109, and 18×109, respectively, compared with the SegNet, PSPNet, U-Net and standard DeepLabv3+. The frame rate of image detection increased by 15.3, 11.7, 11.4, and 9fps, respectively. The coefficient of determination reached 0.981 in the regression analysis with the manually measured root length. The 80% of potato roots were distributed in the soil layers of 0-20, 0-30, 0-40, and 0-30cm, respectively, during the seedling, tuber formation, tuber bulking, and starch accumulation stage. The finding can provide a theoretical basis for the high-yield and high-efficiency cultivation techniques of potato in the northern Yinshan Mountain in Inner Mongolia of China. [ABSTRACT FROM AUTHOR]

Infrared small target detection is critical in remote sensing, military, and other fields. However, the low resolution of most infrared images and the lack of texture and detailed information could cause the target to be lost in a relatively noisy background. Therefore, in recent years, researchers have paid particular attention to the problem of small infrared target detection. In this paper, we propose a double-layer feature fusion convolutional neural network for infrared small target detection (DLFF), consisting of a simultaneous upsampling two-layer network module and a 'T'-type fusion structure. First, the upsampling double-layer network module shares detection information while synchronizing detection, suppressing the background noise and enhancing the detection of the target. In addition, for the small target detection task, since the direct fusion of shallow spatial information and deep semantic information may lose only some small target features, we propose a 'T'-type fusion structure to solve this problem. Furthermore, we collate an infrared small target dataset (MDFA_SIRIST) and design a pre-processing method for pre-detection images. The experimental results show that our network outperforms the other six state-of-the-art methods in combined evaluation metrics ( F 1 -score) and mean intersection ratio (mIou). [ABSTRACT FROM AUTHOR]

Pooling layers are essential building blocks of convolutional neural networks (CNNs), to reduce computational overhead and increase the receptive fields of proceeding convolutional operations. Their goal is to produce downsampled volumes that closely resemble the input volume while, ideally, also being computationally and memory efficient. Meeting both these requirements remains a challenge. To this end, we propose an adaptive and exponentially weighted pooling method: adaPool. Our method learns a regional-specific fusion of two sets of pooling kernels that are based on the exponent of the Dice-Sørensen coefficient and the exponential maximum, respectively. AdaPool improves the preservation of detail on a range of tasks including image and video classification and object detection. A key property of adaPool is its bidirectional nature. In contrast to common pooling methods, the learned weights can also be used to upsample activation maps. We term this method adaUnPool. We evaluate adaUnPool on image and video super-resolution and frame interpolation. For benchmarking, we introduce Inter4K, a novel high-quality, high frame-rate video dataset. Our experiments demonstrate that adaPool systematically achieves better results across tasks and backbones, while introducing a minor additional computational and memory overhead. [ABSTRACT FROM AUTHOR]

High resolution remains a primary goal in the advancement of synthetic aperture radar (SAR) technology. The backprojection (BP) algorithm, which does not introduce any approximation throughout the imaging process, is broadly applicable and effectively meets the demands for high-resolution imaging. Nonetheless, the BP algorithm necessitates substantial interpolation during point-by-point processing, and the precision and effectiveness of current interpolation methods limit the imaging performance of the BP algorithm. This paper proposes a TSU-ICSI (Time-shift Upsampling-Improved Cubic Spline Interpolation) interpolation method that integrates time-shift upsampling with improved cubic spline interpolation. This method is applied to the BP algorithm and presents an efficient implementation method in conjunction with the GPU architecture. TSU-ICSI not only maintains the accuracy of BP imaging processing but also significantly boosts performance. The effectiveness of the BP algorithm based on TSU-ICSI is confirmed through simulation experiments and by processing measured data collected from both airborne SAR and spaceborne SAR.

In clinical diagnosis, an effective classification of ovarian carcinoma types is highly essential to avoid the number of deaths worldwide. For this reason, deep convolutional neural network (DCNN) has been designed to classify ovarian carcinoma previously. Then, insufficiency of a dataset was handled by augmenting the training samples using deep semi-supervised generative learning (DSSGL). But, these augmented images directly fed to the DCNN without segmentation causes improper classification of ovarian carcinoma in a significant regions. Also, its computation burden is high. Hence in this article, an enhanced U-Net (EUNet) is proposed as a segmentation module with the DSSGL-DCNN framework for enhancing the accuracy of classifying ovarian carcinoma. This EUNet comprises different units: the inception-residual (IR) unit, the dense-inception (DI) unit, the downsampling unit and the upsampling unit to create the feature-level segmented maps for a given CT scan. But, raising the expansion ratio in the DI unit will provide several variables which make the framework more complex and slower to train. So, the feature-level probability map is also generated which is thresholded to binary and fused with the feature-level segmented maps to create the discriminative segmented sample. In ovarian carcinoma classification, the training CT images are first augmented by the DSSGL method and given to the EUNet. The resultant segmented images from EUNet are fed to the fused structure-based DCNN for categorizing the types of ovarian carcinomas effectively. Finally, the testing outcomes reveal that the DSSGL-EUNet-DCNN attains 91.63 % of accuracy for ovarian carcinoma categorization, whereas existing MLR, GoogleNet, DHL, 2-level DTEL and DSSGL-DCNN achieve 80.24 %, 82.39 %, 85.51 %, 87.76 %, and 88.98 % respectively. [ABSTRACT FROM AUTHOR]

In a foggy traffic environment, the vision sensor signal of intelligent vehicles will be distorted, the outline of obstacles will become blurred, and the color information in the traffic road will be missing. To solve this problem, four ultra-fast defogging strategies in a traffic environment are proposed for the first time. Through experiments, it is found that the performance of Fast Defogging Strategy 3 is more suitable for fast defogging in a traffic environment. This strategy reduces the original foggy picture by 256 times via bilinear interpolation, and the defogging is processed via the dark channel prior algorithm. Then, the image after fog removal is processed via 4-time upsampling and Gaussian transform. Compared with the original dark channel prior algorithm, the image edge is clearer, and the color information is enhanced. The fast defogging strategy and the original dark channel prior algorithm can reduce the defogging time by 83.93–84.92%. Then, the image after fog removal is inputted into the YOLOv4, YOLOv5, YOLOv6, and YOLOv7 target detection algorithms for detection and verification. It is proven that the image after fog removal can effectively detect vehicles and pedestrians in a complex traffic environment. The experimental results show that the fast defogging strategy is suitable for fast defogging in a traffic environment.

The projection of a point cloud onto a 2D camera image is relevant in the case of various image analysis and enhancement tasks, e.g., (i) in multimodal image processing for data fusion, (ii) in robotic applications and in scene analysis, and (iii) for deep neural networks to generate real datasets with ground truth. The challenges of the current single-shot projection methods, such as simple state-of-the-art projection, conventional, polygon, and deep learning-based upsampling methods or closed source SDK functions of low-cost depth cameras, have been identified. We developed a new way to project point clouds onto a dense, accurate 2D raster image, called Triangle-Mesh-Rasterization-Projection (TMRP). The only gaps that the 2D image still contains with our method are valid gaps that result from the physical limits of the capturing cameras. Dense accuracy is achieved by simultaneously using the 2D neighborhood information (rx,ry) of the 3D coordinates in addition to the points P(X,Y,V). In this way, a fast triangulation interpolation can be performed. The interpolation weights are determined using sub-triangles. Compared to single-shot methods, our algorithm is able to solve the following challenges. This means that: (1) no false gaps or false neighborhoods are generated, (2) the density is XYZ independent, and (3) ambiguities are eliminated. Our TMRP method is also open source, freely available on GitHub, and can be applied to almost any sensor or modality. We also demonstrate the usefulness of our method with four use cases by using the KITTI-2012 dataset or sensors with different modalities. Our goal is to improve recognition tasks and processing optimization in the perception of transparent objects for robotic manufacturing processes.

With the development of deep learning, semantic segmentation has received considerable attention within the robotics community. For semantic segmentation to be applied to mobile robots or autonomous vehicles, real-time processing is essential. In this article, a new real-time semantic segmentation network, called the adjacent feature propagation network (AFPNet), is proposed to achieve high performance and fast inference. AFPNet executes in real time on a commercial embedded GPU. The network includes two new modules. The local memory module (LMM) is the first; it improves the upsampling accuracy by propagating the high-level features to the adjacent grids. The cascaded pyramid pooling module (CPPM) is the second; it reduces computational time by changing the structure of the pyramid pooling module. Using these two modules, the proposed AFPNet achieved 76.4% mean intersection-over-union on the Cityscapes test dataset, outperforming other real-time semantic segmentation networks. Furthermore, AFPNet was successfully deployed on an embedded board Jetson AGX Xavier and applied to the real-world navigation of a mobile robot, proving that AFPNet can be effectively used in a variety of real-time applications. [ABSTRACT FROM AUTHOR]

Point cloud upsampling is vital for the quality of the mesh in three-dimensional reconstruction. Recent research on point cloud upsampling has achieved great success due to the development of deep learning. However, the existing methods regard point cloud upsampling of different scale factors as independent tasks. Thus, the methods need to train a specific model for each scale factor, which is both inefficient and impractical for storage and computation in real applications. To address this limitation, in this article, we propose a novel method called “Meta-PU” to first support point cloud upsampling of arbitrary scale factors with a single model. In the Meta-PU method, besides the backbone network consisting of residual graph convolution (RGC) blocks, a meta-subnetwork is learned to adjust the weights of the RGC blocks dynamically, and a farthest sampling block is adopted to sample different numbers of points. Together, these two blocks enable our Meta-PU to continuously upsample the point cloud with arbitrary scale factors by using only a single model. In addition, the experiments reveal that training on multiple scales simultaneously is beneficial to each other. Thus, Meta-PU even outperforms the existing methods trained for a specific scale factor only. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Computer Engineering & Applications is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In recent years, the highly boosting development in e-commerce technologies made it possible for people to select the most desirable items from shops and stores worldwide while being at home. Credit card frauds transactions are common nowadays because of online payments. Online transactions are the root cause of fraudulent credit card activity, bringing enormous financial losses. Financial institutions must install an automatic deterrent mechanism to check these fraudulent actions. The fraudulent transactions do not follow a specific pattern and continuously change their shape and behavior. This paper aims to use ensemble learning with supervised Machine Learning (ML) models to predict the occurrence of fraud transactions. The experimental study has been evaluated on the opensource Kaggle credit card fraud detection dataset. The performance of the proposed model is measured in terms of accuracy score, confusion matrix, and classification report. The results were state-of-the-art using the voting ensemble learning technique shows that it can be get the best results using PCA with 100.0% accuracy, 97.3% precision, 73.5% recall, and 83.7% f1-score against other ML classifiers. [ABSTRACT FROM AUTHOR]

Today, target detection has an indispensable application in various fields. Infrared small-target detection, as a branch of target detection, can improve the perception capability of autonomous systems, and it has good application prospects in infrared alarm, automatic driving and other fields. There are many well-established algorithms that perform well in infrared small-target detection. Nevertheless, the current algorithms cannot achieve the expected detection effect in complex environments, such as background clutter, noise inundation or very small targets. We have designed an image enhancement-based detection algorithm to solve both problems through detail enhancement and target expansion. This method first enhances the mutation information, detail and edge information of the image and then improves the contrast between the target edge and the adjacent pixels to make the target more prominent. The enhancement improves the robustness of detection with background clutter or noise-flooded scenes. Moreover, bicubic interpolation is used on the input image, and the target pixels are expanded with upsampling, which enhances the detection effectiveness for tiny targets. From the results of qualitative and quantitative experiments, the algorithm proposed in this paper outperforms the existing work on various evaluation indicators. [ABSTRACT FROM AUTHOR]

Due to the increased popularity of augmented and virtual reality experiences, the interest in capturing high-resolution real-world point clouds has never been higher. Loss of details and irregularities in point cloud geometry can occur during the capturing, processing, and compression pipeline. It is essential to address these challenges by being able to upsample a low Level-of-Detail (LoD) point cloud into a high LoD point cloud. Current upsampling methods suffer from several weaknesses in handling point cloud upsampling, especially in dense real-world photo-realistic point clouds. In this paper, we present a novel geometry upsampling technique, PU-Dense, which can process a diverse set of point clouds including synthetic mesh-based point clouds, real-world high-resolution point clouds, real-world indoor LiDAR scanned objects, as well as outdoor dynamically acquired LiDAR-based point clouds. PU-Dense employs a 3D multiscale architecture using sparse convolutional networks that hierarchically reconstruct an upsampled point cloud geometry via progressive rescaling and multiscale feature extraction. The framework employs a UNet type architecture that downscales the point cloud to a bottleneck and then upscales it to a higher level-of-detail (LoD) point cloud. PU-Dense introduces a novel Feature Extraction Unit that incorporates multiscale spatial learning by employing filters at multiple sampling rates and receptive fields. The architecture is memory efficient and is driven by a binary voxel occupancy classification loss that allows it to process high-resolution dense point clouds with millions of points during inference time. Qualitative and quantitative experimental results show that our method significantly outperforms the state-of-the-art approaches by a large margin while having much lower inference time complexity. We further test our dataset on high-resolution photo-realistic datasets. In addition, our method can handle noisy data well. We further show that our approach is memory efficient compared to the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The task of point cloud upsampling aims to acquire dense and uniform point sets from sparse and irregular point sets. Although significant progress has been made with deep learning models, state-of-the-art methods require ground-truth dense point sets as the supervision, which makes them limited to be trained under synthetic paired training data and not suitable to be under real-scanned sparse data. However, it is expensive and tedious to obtain large numbers of paired sparse-dense point sets as supervision from real-scanned sparse data. To address this problem, we propose a self-supervised point cloud upsampling network, named SPU-Net, to capture the inherent upsampling patterns of points lying on the underlying object surface. Specifically, we propose a coarse-to-fine reconstruction framework, which contains two main components: point feature extraction and point feature expansion, respectively. In the point feature extraction, we integrate the self-attention module with the graph convolution network (GCN) to capture context information inside and among local regions simultaneously. In the point feature expansion, we introduce a hierarchically learnable folding strategy to generate upsampled point sets with learnable 2D grids. Moreover, to further optimize the noisy points in the generated point sets, we propose a novel self-projection optimization associated with uniform and reconstruction terms as a joint loss to facilitate the self-supervised point cloud upsampling. We conduct various experiments on both synthetic and real-scanned datasets, and the results demonstrate that we achieve comparable performances to state-of-the-art supervised methods. [ABSTRACT FROM AUTHOR]

To solve the problems of low recognition rate and high miss rate of small object detection in the current target detection task, an improved YOLOv3 algorithm based on gated channel attention mechanism (EGCA) and adaptive upsampling module is proposed. Firstly, darknet-53 is used as the backbone network for image basic feature extraction. Secondly, the adaptive upsampling module is introduced to expand the low-resolution convolution feature map, which effectively enhances the fusion effect of different scale convolution feature map. Finally, EGCA attention mechanism is added before the three scale channels output the prediction results to improve the feature expression and detection ability of the network to small objects. The improved algorithm is tested on the public data set RSOD (remote sensing object detection), the accuracy of small object detection is improved by 8.2 percentage points, which is the most significant. The average accuracy AP value reaches 56.3%, which is 7.9 percentage points higher than the original algorithm. Experimental results show that the improved algorithm has better small object detection ability than the traditional YOLOv3 algorithm and other algorithms. [ABSTRACT FROM AUTHOR]