Search

In semantic segmentation research, spatial information and receptive fields are essential. However, currently, most algorithms focus on acquiring semantic information and lose a significant amount of spatial information, leading to a significant decrease in accuracy despite improving real-time inference speed. This paper proposes a new method to address this issue. Specifically, we have designed a new module (HFRM) that combines channel attention and spatial attention to retrieve the spatial information lost during downsampling and enhance object classification accuracy. Regarding fusing spatial and semantic information, we have designed a new module (HFFM) to merge features of two different levels more effectively and capture a larger receptive field through an attention mechanism. Additionally, edge detection methods have been incorporated to enhance the extraction of boundary information. Experimental results demonstrate that for an input size of 512 × 1024, our proposed method achieves 73.6% mIoU at 176 frames per second (FPS) on the Cityscapes dataset and 70.0% mIoU at 146 FPS on Camvid. Compared to existing networks, our Model achieves faster inference speed while maintaining accuracy, enhancing its practicality. [ABSTRACT FROM AUTHOR]

We present a Spiking Neural Network (SNN) model that incorporates learnable synaptic delays through two approaches: per-synapse delay learning via Dilated Convolutions with Learnable Spacings (DCLS) and a dynamic pruning strategy that also serves as a form of delay learning. In the latter approach, the network dynamically selects and prunes connections, optimizing the delays in sparse connectivity settings. We evaluate both approaches on the Raw Heidelberg Digits keyword spotting benchmark using Backpropagation Through Time with surrogate gradients. Our analysis of the spatio-temporal structure of synaptic interactions reveals that, after training, excitation and inhibition group together in space and time. Notably, the dynamic pruning approach, which employs DEEP R for connection removal and RigL for reconnection, not only preserves these spatio-temporal patterns but outperforms per-synapse delay learning in sparse networks. Our results demonstrate the potential of combining delay learning with dynamic pruning to develop efficient SNN models for temporal data processing. Moreover, the preservation of spatio-temporal dynamics throughout pruning and rewiring highlights the robustness of these features, providing a solid foundation for future neuromorphic computing applications. [ABSTRACT FROM AUTHOR]

Semantic segmentation, a task of assigning class labels to each pixel in an image, has found applications in various real‐world scenarios, including autonomous driving and scene understanding. However, its widespread use is hindered by the high computational burden. In this paper, we propose an efficient semantic segmentation method based on Feature Cascade Fusion Network (FCFNet) to address this challenge. FCFNet utilizes a dual‐path framework comprising the Spatial Information Path (SIP) and the Context Information Path (CIP). SIP is a shallow structure that captures the local dependencies of each pixel to improve the accuracy of detailed segmentation. CIP is the main branch with a deeper structure that captures sufficient contextual information from input features. Moreover, we design an Efficient Receptive Field Module (ERFM) to enlarge the receptive field in the SIP. Meanwhile, Attention Shuffled Refinement Module is used to refine feature maps from different stages. Finally, we present an Attention‐Guided Fusion Module to fuse the low‐ and high‐level feature maps effectively. Experimental results show that our proposed FCFNet achieves 70.7% mean intersection over union (mIoU) on the Cityscapes data set and 68.1% mIoU on the CamVid data set, respectively, with inference speeds of 110 and 100 frames per second (FPS), respectively. Additionally, we evaluated FCFNet on the Nvidia Jetson Xavier embedded device, which demonstrated competitive performance while significantly reducing power consumption. [ABSTRACT FROM AUTHOR]

Remote sensing images possess abundant texture features and significant autocorrelation. However, the extensive network parameters and high computational demands of current super-resolution (SR) methods make them challenging to implement on mobile devices. This work proposes a lightweight model named the Sequential Multi-Scale Feature Enhancement Network (SMFEN) that address the issue on single remote sensing image super-resolution. Our sequential structure allows for a larger receptive field (RF) with minimal parameters, which can gradually build complex high-level multi-scale feature representations from simple low-level features, realizing the feature extraction process from concrete to abstract. In addition, we design a high-frequency multi-scale attention block which use multi-scale high-frequency details to facilitate the fusion of contextual information across different scales and effectively recover texture and edge information. Comprehensive experimental results demonstrate that our SMFEN network outperforms the latest lightweight SR methods. [ABSTRACT FROM AUTHOR]

Key points A fundamental question in sensory neuroscience revolves around how neurons represent complex visual stimuli. In mammalian primary visual cortex (V1), neurons decode intricate visual features to identify objects, with most being selective for edge orientation, but with half of those also developing invariance to edge position within their receptive fields. Position invariance allows cells to continue to code an edge even when it moves around. Combining feature selectivity and invariance is integral to successful object recognition. Considering the marsupial–eutherian divergence 160 million years ago, we explored whether feature selectivity and invariance was similar in marsupials and eutherians. We recovered the spatial filters and non‐linear processing characteristics of the receptive fields of neurons in wallaby V1 and compared them with previous results from cat cortex. We stimulated the neurons in V1 with white Gaussian noise and analysed responses using the non‐linear input model. Wallabies exhibit the same high percentage of orientation selective neurons as cats. However, in wallabies we observed a notably higher prevalence of neurons with three or more filters compared to cats. We show that having three or more filters substantially increases phase invariance in the V1s of both species, but that wallaby V1 accentuates this feature, suggesting that the species condenses more processing into the earliest cortical stage. These findings suggest that evolution has led to more than one solution to the problem of creating complex visual processing strategies. Previous studies have shown that the primary visual cortex (V1) in mammals is essential for processing complex visual stimuli, with neurons displaying selectivity for edge orientation and position. This research explores whether the visual processing mechanisms in marsupials, such as wallabies, are similar to those in eutherian mammals (e.g. cats). The study found that wallabies have a higher prevalence of neurons with multiple spatial filters in V1, indicating more complex visual processing. Using a non‐linear input model, we demonstrated that neurons with three or more filters increase phase invariance. These findings suggest that marsupials and eutherian mammals have evolved similar strategies for visual processing, but marsupials have condensed more capacity to build phase invariance into the first step in the cortical pathway. [ABSTRACT FROM AUTHOR]

Vehicle trajectory prediction plays a crucial role in the control and safety warning of autonomous vehicles. Existing methods often depend on costly high definition (HD) maps for generating trajectories to fit their scenarios, or involve inefficient aggregation of local point clouds into voxels. Therefore, an end-to-end vehicle trajectory prediction method (PillarVTP) is proposed based on local point cloud aggregation and receptive field expansion. Firstly, we construct a novel pillar-based object detection network, introducing SPPCSPC which uses max pooling layers with multiple kernel sizes on a single feature level as the neck for extracting multi-scale features, and improving ResNet-18 by adding a depth stage to expand the receptive field at multiple levels. Then, we present performing feature upsampling to improve performance before predicting vehicle positions. And a shallow convolutional network is utilized to implement the future feature learning network, which learns future features from the previous features for predicting vehicle positions in future frames. Subsequently, the positions of vehicles are matched greedily from future frames to the current frame, and the matched future trajectories are associated with the vehicles detected in the current frame. Finally, the proposed PillarVTP is evaluated on the nuScenes and Argoverse 1 datasets. Experimental results demonstrate that PillarVTP outperforms recent end-to-end prediction method based on point cloud data, FutureDet, by 3.4% and surpasses traditional multi-stage method, Trajectron + + , by 13.7%. Furthermore, PillarVTP shows good robustness under various weather conditions. [ABSTRACT FROM AUTHOR]

Underwater object detection is vital for diverse applications, from studies in marine biology to underwater robotics. However, underwater environments pose unique challenges, including reduced visibility due to color distortion, light attenuation, and complex backgrounds. Traditional computer vision methods have limitations, prompting the implementation of deep learning, for underwater object detection. Despite progress, challenges persist, such as visual degradation, scale variations, diverse marine species, and complex backgrounds. To address these issues, we propose Feature-Adaptive FPN with Multiscale Context Integration (FA-FPN-MCI), a novel deep-learning algorithm aimed at enhancing both detection and domain generalization performance. We integrate the Style Normalization and Restitution (SNR) module for domain generalization, Receptive Field Blocks (RFBs) for fine-grained detail capture, and a twin-branch Global Context Module (TBGCM) for multiscale context information. We enhance lateral connections within the Feature Pyramid Network (FPN) with deformable convolution. Experimental outcome reveal that the proposed method attains mean average precision of 84.2%. Additionally, other performance metrics were evaluated, and outperforming all other methods used for comparison. [ABSTRACT FROM AUTHOR]

Brain tumors are serious conditions, which can cause great trauma to patients, endangering their health and even leading to disability or death. Therefore, accurate preoperative diagnosis is particularly important. Accurate brain tumor segmentation based on deep learning plays an important role in the preoperative treatment planning process and has achieved good performance. However, one of the challenges involved is an insufficient ability to extract features with a large receptive field in encoder layers and guide the selection of deep semantic information in decoder layers. We propose small kernel convolution with an effective receptive field network (SCENet) based on UNet, which involves a small kernel convolution with effective receptive field shuffle module (SCER) and a channel spatial attention module (CSAM). The SCER module utilizes the inherent properties of stacking convolution to obtain effectively receptive fields and improve the features with a large receptive field extraction ability. CSAM of decoder layers can preserve more detailed features to capture clearer contours of the segmented image by calculating the weights of channels and spaces. An ASPP module is introduced to the bottleneck layer to enlarge the receptive field and can capture multi-scale detailed features. Furthermore, a large number of experiments were performed to evaluate the performance of our model on the BraTS2021 dataset. The SCENet achieved dice coefficient scores of 91.67%, 87.70%, and 83.35% for whole tumor (WT), tumor core (TC), and enhancing tumor (ET), respectively. The results show that the proposed model achieves the state-of-the-art performance compared with more than twelve benchmarks. [ABSTRACT FROM AUTHOR]

Surface defect detection plays a crucial role in ensuring the quality standards of hot-rolled steel strips. To meet the demands for high precision and real-time performance in industrial defect detection, this paper introduces an improved one-stage detector based on YOLOv5s, named GLF-NET, that focuses on a good balance between speed and precision. Firstly, an Attention Augmented Module (AAM) is proposed and used in the backbone, with the aim of minimizing the loss of semantic and location information of defects during the process of feature extraction. Secondly, to enrich the model’s capacity of representing multi-scale features of defects, an innovative Global and Local Dynamic Feature Fusion (GLF) module is designed and plugged into the top-down FPN part of the neck, bridging the semantic gap between different feature layers and enabling the model to adaptively select features for fusion. Additionally, a novel Receptive Field Augmented Module (RFA) is proposed and integrated into the bottom-up PAN structure of the neck, enhancing the detector’s ability of perceiving defects with irregular shapes and large aspect ratios. Extensive experimental results on the NEU-DET steel strip surface defect dataset demonstrate that GLF-NET obtains an impressive mAP value of 79.2%, exceeding YOLOv5s by 4.2%. Furthermore, with an impressive detection speed of 95 Frames Per Second (FPS), GLF-NET not only meets the real-time demands of industrial defect detection but also demonstrates exceptional capabilities in defect detection. Code is available at https://github.com/MYF1124/GLF-NET.

IntroductionIn primary auditory cortex (A1), nicotinic acetylcholine receptors (nAChRs) containing α2 subunits are expressed in layer 5 Martinotti cells (MCs)-inhibitory interneurons that send a main axon to superficial layers to inhibit distal apical dendrites of pyramidal cells (PCs). MCs also contact interneurons in supragranular layers that, in turn, inhibit PCs. Thus, MCs may regulate PCs via inhibition and disinhibition, respectively, of distal and proximal apical dendrites. Auditory inputs to PCs include thalamocortical inputs to middle layers relaying information about characteristic frequency (CF) and near-CF stimuli, and intracortical long-distance ("horizontal") projections to multiple layers carrying information about spectrally distant ("nonCF") stimuli. CF and nonCF inputs integrate to create broad frequency receptive fields (RFs). Systemic administration of nicotine activates nAChRs to "sharpen" RFs-to increase gain within a narrowed RF-resulting in enhanced responses to CF stimuli and reduced responses to nonCF stimuli. While nicotinic mechanisms to increase gain have been identified, the mechanism underlying RF narrowing is unknown.MethodsHere, we examine the role of α2 nAChRs in mice with α2 nAChR-expressing neurons labeled fluorescently, and in mice with α2 nAChRs genetically deleted.ResultsThe distribution of fluorescent neurons in auditory cortex was consistent with previous studies demonstrating α2 nAChRs in layer 5 MCs, including nonpyramidal somata in layer 5 and dense processes in layer 1. We also observed label in subcortical auditory regions, including processes, but no somata, in the medial geniculate body, and both fibers and somata in the inferior colliculus. Using electrophysiological (current-source density) recordings in α2 nAChR knock-out mice, we found that systemic nicotine failed to enhance CF-evoked inputs to layer 4, suggesting a role for subcortical α2 nAChRs, and failed to reduce nonCF-evoked responses, suggesting that α2 nAChRs regulate horizontal projections to produce RF narrowing.DiscussionThe results support the hypothesis that α2 nAChRs function to simultaneously enhance RF gain and narrow RF breadth in A1. Notably, a similar neural circuit may recur throughout cortex and hippocampus, suggesting widespread conserved functions regulated by α2 nAChRs.

Abstract This paper addresses the challenge of low detection accuracy of grape clusters caused by scale differences, illumination changes, and occlusion in realistic and complex scenes. We propose a multi-scale feature fusion and augmentation YOLOv7 network to enhance the detection accuracy of grape clusters across variable environments. First, we design a Multi-Scale Feature Extraction Module (MSFEM) to enhance feature extraction for small-scale targets. Second, we propose the Receptive Field Augmentation Module (RFAM), which uses dilated convolution to expand the receptive field and enhance the detection accuracy for objects of various scales. Third, we present the Spatial Pyramid Pooling Cross Stage Partial Concatenation Faster (SPPCSPCF) module to fuse multi-scale features, improving accuracy and speeding up model training. Finally, we integrate the Residual Global Attention Mechanism (ResGAM) into the network to better focus on crucial regions and features. Experimental results show that our proposed method achieves a mAP $$_{0.5}$$ 0.5 of 93.29% on the GrappoliV2 dataset, an improvement of 5.39% over YOLOv7. Additionally, our method increases Precision, Recall, and F1 score by 2.83%, 3.49%, and 0.07, respectively. Compared to state-of-the-art detection methods, our approach demonstrates superior detection performance and adaptability to various environments for detecting grape clusters.

A key feature of the receptive field of neurons in the visual system is their centre–surround antagonism, whereby the centre and the surround exhibit responses of opposite polarity. This organization is thought to enhance visual acuity, but whether and how such antagonism plays a role in more complex processing remains poorly understood. Here, we investigate the role of centre and surround receptive fields in retinal direction selectivity by exposing posterior‐preferring On–Off direction‐selective ganglion cells (pDSGCs) to adaptive light and recording their response to globally moving objects. We reveal that light adaptation leads to surround expansion in pDSGCs. The pDSGCs maintain their original directional tuning in the centre receptive field, but present the oppositely tuned response in their surround. Notably, although inhibition is the main substrate for retinal direction selectivity, we found that following light adaptation, both the centre‐ and surround‐mediated responses originate from directionally tuned excitatory inputs. Multi‐electrode array recordings show similar oppositely tuned responses in other DSGC subtypes. Together, these data attribute a new role for excitation in the direction‐selective circuit. This excitation carries an antagonistic centre–surround property, possibly designed to sharpen the detection of motion direction in the retina. Key points: Receptive fields of direction‐selective retinal ganglion cells expand asymmetrically following light adaptation.The increase in the surround receptive field generates a delayed spiking phase that is tuned to the null direction and is mediated by excitation.Following light adaptation, excitation rules the computation in the centre receptive field and is tuned to the preferred direction.GABAergic and glycinergic inputs modulate the null‐tuned delayed response differentially.Null‐tuned delayed spiking phases can be detected in all types of direction‐selective retinal ganglion cells.Light adaptation exposes a hidden directional excitation in the circuit, which is tuned to opposite directions in the centre and surround receptive fields. [ABSTRACT FROM AUTHOR]

Object detection usually adopts multi-scale fusion to enrich the information of the object, and the Feature Pyramid Network (FPN) is a common method for multi-scale fusion. However, traditional fusion methods such as FPN cause information loss when fusing high-level feature maps with low-level feature maps. To solve these problems, we propose a simple but effective cross-scale fusion method that fully uses the information of multi-scale feature maps. In addition, to better utilize the multi-scale contextual information, we designed the Selective Information Enhancement (SIE) module. The SIE dynamically selects information at more important scales for objects of different size and fuse the selected information with feature maps for information enhancement. Apply our method to Single Shot Multibox Detector (SSD) and propose a Cross-Scale Information Enhancement Single Shot Multibox Detector (CESSD). The CESSD improves the object detection capability of SSD models by fusing multi-scale features and selectively enhancing feature map information. To evaluate the effectiveness of the model, we validated it on the Pascal VOC2007 test set for 300 × 300 inputs, and the mean Average Precision (mAP) of CESSD reached 79.8%. [ABSTRACT FROM AUTHOR]

This paper addresses the challenge of low detection accuracy of grape clusters caused by scale differences, illumination changes, and occlusion in realistic and complex scenes. We propose a multi-scale feature fusion and augmentation YOLOv7 network to enhance the detection accuracy of grape clusters across variable environments. First, we design a Multi-Scale Feature Extraction Module (MSFEM) to enhance feature extraction for small-scale targets. Second, we propose the Receptive Field Augmentation Module (RFAM), which uses dilated convolution to expand the receptive field and enhance the detection accuracy for objects of various scales. Third, we present the Spatial Pyramid Pooling Cross Stage Partial Concatenation Faster (SPPCSPCF) module to fuse multi-scale features, improving accuracy and speeding up model training. Finally, we integrate the Residual Global Attention Mechanism (ResGAM) into the network to better focus on crucial regions and features. Experimental results show that our proposed method achieves a mAP 0.5 of 93.29% on the GrappoliV2 dataset, an improvement of 5.39% over YOLOv7. Additionally, our method increases Precision, Recall, and F1 score by 2.83%, 3.49%, and 0.07, respectively. Compared to state-of-the-art detection methods, our approach demonstrates superior detection performance and adaptability to various environments for detecting grape clusters. [ABSTRACT FROM AUTHOR]

Object detection in aerial images has had a broader range of applications in the past few years. Unlike the targets in the images of horizontal shooting, targets in aerial photos generally have arbitrary orientation, multi-scale, and a high aspect ratio. Existing methods often employ a classification backbone network to extract translation-equivariant features (TEFs) and utilize many predefined anchors to handle objects with diverse appearance variations. However, they encounter misalignment at three levels, spatial, feature, and task, during different detection stages. In this study, we propose a model called the Staged Adaptive Alignment Detector (SAADet) to solve these challenges. This method utilizes a Spatial Selection Adaptive Network (SSANet) to achieve spatial alignment of the convolution receptive field to the scale of the object by using a convolution sequence with an increasing dilation rate to capture the spatial context information of different ranges and evaluating this information through model dynamic weighting. After correcting the preset horizontal anchor to an oriented anchor, feature alignment is achieved through the alignment convolution guided by oriented anchor to align the backbone features with the object's orientation. The decoupling of features using the Active Rotating Filter is performed to mitigate inconsistencies due to the sharing of backbone features in regression and classification tasks to accomplish task alignment. The experimental results show that SAADet achieves equilibrium in speed and accuracy on two aerial image datasets, HRSC2016 and UCAS-AOD. [ABSTRACT FROM AUTHOR]

In recent years, forest fires have been occurring frequently around the globe, affected by extreme weather and dry climate, causing serious economic losses and environmental pollution. In this context, timely detection of forest fire smoke is crucial for realizing real-time early warning of fires. However, fire and smoke from forest fires can spread to cover large areas and may affect distant areas. In this paper, a lightweight joint forest fire and smoke detection network, SmokeFireNet, is proposed, which employs ShuffleNetV2 as the backbone for efficient feature extraction, effectively addressing the computational efficiency challenges of traditional methods. To integrate multi-scale information and enhance the semantic feature extraction capability, a feature pyramid network (FPN) and path aggregation network (PAN) are introduced in this paper. In addition, the FPN network is optimized by a lightweight DySample upsampling operator. The model also incorporates efficient channel attention (ECA), which can pay more attention to the detection of forest fires and smoke regions while suppressing irrelevant features. Finally, by embedding the receptive field block (RFB), the model further improves its ability to understand contextual information and capture detailed features of fire and smoke, thus improving the overall detection accuracy. The experimental results show that SmokeFireNet is better than other mainstream target detection algorithms in terms of average APall of 86.2%, FPS of 114, and GFLOPs of 8.4, and provides effective technical support for forest fire prevention work in terms of average precision, frame rate, and computational complexity. In the future, the SmokeFireNet model is expected to play a greater role in the field of forest fire prevention and make a greater contribution to the protection of forest resources and the ecological environment. [ABSTRACT FROM AUTHOR]

Brain tumors are frequently classified with high accuracy using convolutional neural networks (CNNs) to better comprehend the spatial connections among pixels in complex pictures. Due to their tiny receptive fields, the majority of deep convolutional neural network (DCNN)-based techniques overfit and are unable to extract global context information from more significant regions. While dilated convolution retains data resolution at the output layer and increases the receptive field without adding computation, stacking several dilated convolutions has the drawback of producing a grid effect. This research suggests a dilated parallel deep convolutional neural network (PDCNN) architecture that preserves a wide receptive field in order to handle gridding artifacts and extract both coarse and fine features from the images. This article applies multiple preprocessing strategies to the input MRI images used to train the model. By contrasting various dilation rates, the global path uses a low dilation rate (2,1,1), while the local path uses a high dilation rate (4,2,1) for decremental even numbers to tackle gridding artifacts and to extract both coarse and fine features from the two parallel paths. Using three different types of MRI datasets, the suggested dilated PDCNN with the average ensemble method performs best. The accuracy achieved for the multiclass Kaggle dataset-III, Figshare dataset-II, and binary tumor identification dataset-I is 98.35%, 98.13%, and 98.67%, respectively. In comparison to state-of-the-art techniques, the suggested structure improves results by extracting both fine and coarse features, making it efficient. [ABSTRACT FROM AUTHOR]

We present a Spiking Neural Network (SNN) model that incorporates learnable synaptic delays through two approaches: per-synapse delay learning via Dilated Convolutions with Learnable Spacings (DCLS) and a dynamic pruning strategy that also serves as a form of delay learning. In the latter approach, the network dynamically selects and prunes connections, optimizing the delays in sparse connectivity settings. We evaluate both approaches on the Raw Heidelberg Digits keyword spotting benchmark using Backpropagation Through Time with surrogate gradients. Our analysis of the spatio-temporal structure of synaptic interactions reveals that, after training, excitation and inhibition group together in space and time. Notably, the dynamic pruning approach, which employs DEEP R for connection removal and RigL for reconnection, not only preserves these spatio-temporal patterns but outperforms per-synapse delay learning in sparse networks. Our results demonstrate the potential of combining delay learning with dynamic pruning to develop efficient SNN models for temporal data processing. Moreover, the preservation of spatio-temporal dynamics throughout pruning and rewiring highlights the robustness of these features, providing a solid foundation for future neuromorphic computing applications.

Brain tumors are frequently classified with high accuracy using convolutional neural networks (CNNs) to better comprehend the spatial connections among pixels in complex pictures. Due to their tiny receptive fields, the majority of deep convolutional neural network (DCNN)-based techniques overfit and are unable to extract global context information from more significant regions. While dilated convolution retains data resolution at the output layer and increases the receptive field without adding computation, stacking several dilated convolutions has the drawback of producing a grid effect. This research suggests a dilated parallel deep convolutional neural network (PDCNN) architecture that preserves a wide receptive field in order to handle gridding artifacts and extract both coarse and fine features from the images. This article applies multiple preprocessing strategies to the input MRI images used to train the model. By contrasting various dilation rates, the global path uses a low dilation rate (2,1,1), while the local path uses a high dilation rate (4,2,1) for decremental even numbers to tackle gridding artifacts and to extract both coarse and fine features from the two parallel paths. Using three different types of MRI datasets, the suggested dilated PDCNN with the average ensemble method performs best. The accuracy achieved for the multiclass Kaggle dataset-III, Figshare dataset-II, and binary tumor identification dataset-I is 98.35%, 98.13%, and 98.67%, respectively. In comparison to state-of-the-art techniques, the suggested structure improves results by extracting both fine and coarse features, making it efficient.

Human pose estimation has important applications in medical diagnosis (such as early diagnosis of autism in children and assisting with the diagnosis of Parkinson's disease), human-computer interaction, animation, and other fields. Currently, many human pose estimation algorithms are based on deep learning. However, most research focuses only on increasing the depth and width of the network model. This approach overlooks that merely enlarging the network's depth and width results in excessive parameterization, without enhancing the model's effective receptive field or its ability to extract multi-scale features. Hence, this paper constructs a network model, named MS-HRNet (Multi-Scale High-Resolution Network), for human pose estimation. Specifically, we propose a more concise and efficient version of HRNet framework as the backbone network of MS-HRNet. This addresses the challenges of HRNet complex structure and large number of parameters that cause training difficulties, and its inadequacy in handling multi-scale information. Additionally, we designed a multi-scale convolutional kernel parallel module named MSBlock (Multi-Scale Block) as the basic block of MS-HRNet. By introducing coordinate attention modules and ASFF (Adaptive Spatial Feature Fusion) modules, the model's ability to extract information is effectively increased, and the issue of feature conflict during the fusion of features with different resolutions is resolved, with only a small increase in the number of model parameters. To evaluate the effectiveness of the proposed model, we conducted comparison experiment and ablation experiments using popular human pose estimation datasets, including COCO2017 and MPII, against multiple existing human pose estimation models.On the COCO 2017 dataset, the number of MS-HRNet parameters are decreased by 41% than the baseline model HRNet, the computational complexity by 59%, and the detection accuracies(mAP) are increased by 2.4 point. [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 the context of power-line scenarios characterized by complex backgrounds and diverse scales and shapes of targets, and addressing issues such as large model parameter sizes, insufficient feature extraction, and the susceptibility to missing small targets in engineering-vehicle detection tasks, a lightweight detection algorithm termed CER-YOLOv5s is firstly proposed. The C3 module was restructured by embedding a lightweight Ghost bottleneck structure and convolutional attention module, enhancing the model's ability to extract key features while reducing computational costs. Secondly, an E-BiFPN feature pyramid network is proposed, utilizing channel attention mechanisms to effectively suppress background noise and enhance the model's focus on important regions. Bidirectional connections were introduced to optimize the feature fusion paths, improving the efficiency of multi-scale feature fusion. At the same time, in the feature fusion part, an ERM (enhanced receptive module) was added to expand the receptive field of shallow feature maps through multiple convolution repetitions, enhancing the global information perception capability in relation to small targets. Lastly, a Soft-DIoU-NMS suppression algorithm is proposed to improve the candidate box selection mechanism, addressing the issue of suboptimal detection of occluded targets. The experimental results indicated that compared with the baseline YOLOv5s algorithm, the improved algorithm reduced parameters and computations by 27.8% and 31.9%, respectively. The mean average precision (mAP) increased by 2.9%, reaching 98.3%. This improvement surpasses recent mainstream algorithms and suggests stronger robustness across various scenarios. The algorithm meets the lightweight requirements for embedded devices in power-line scenarios. [ABSTRACT FROM AUTHOR]

Copyright of Computer Measurement & Control is the property of Magazine Agency of Computer Measurement & Control 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.)

Similar to convolutional neural networks for image processing, existing analysis methods for 3D point clouds often require the designation of a local neighborhood to describe the local features of the point cloud. This local neighborhood is typically manually specified, which makes it impossible for the network to dynamically adjust the receptive field's range. If the range is too large, it tends to overlook local details, and if it is too small, it cannot establish global dependencies. To address this issue, we introduce in this paper a new concept: receptive field space (RFS). With a minor computational cost, we extract features from multiple consecutive receptive field ranges to form this new receptive field space. On this basis, we further propose a receptive field space attention mechanism, enabling the network to adaptively select the most effective receptive field range from RFS, thus equipping the network with the ability to adjust granularity adaptively. Our approach achieved state-of-the-art performance in both point cloud classification, with an overall accuracy (OA) of 94.2%, and part segmentation, achieving an mIoU of 86.0%, demonstrating the effectiveness of our method. [ABSTRACT FROM AUTHOR]

Aiming at the problems that the edge of melanoma image is fuzzy, the contrast with the background is low, and the hair occlusion makes it difficult to segment accurately, this paper proposes a model MSCNet for melanoma segmentation based on U-net frame. Firstly, a multi-scale pyramid fusion module is designed to reconstruct the skip connection and transmit global information to the decoder. Secondly, the contextural information conduction module is innovatively added to the top of the encoder. The module provides different receptive fields for the segmented target by using the hole convolution with different expansion rates, so as to better fuse multi-scale contextural information. In addition, in order to suppress redundant information in the input image and pay more attention to melanoma feature information, global channel attention mechanism is introduced into the decoder. Finally, In order to solve the problem of lesion class imbalance, this paper uses a combined loss function. The algorithm of this paper is verified on ISIC 2017 and ISIC 2018 public datasets. The experimental results indicate that the proposed algorithm has better accuracy for melanoma segmentation compared with other CNN-based image segmentation algorithms. [ABSTRACT FROM AUTHOR]

The currently popular image inpainting methods based on deep neural network typically employ large receptive field feature extractors. However, when restoring local patterns and textures, they often generate artifacts or distorted textures, thus failing to recover the overall semantic and visual structure of the image. To address this issue, this paper proposed a novel image inpainting method, called ORFNet, which combined coarse and fine inpainting by employing an optimized receptive field strategy. Initially, it obtained a coarse inpainting result by using a generative adversarial network with a large receptive field. Subsequently, it used a model with a small receptive field to refine local texture details. Finally, it performed a global refinement inpainting by using an encoder-decoder network based on attention mechanisms. Validation on the CelebA, Paris StreetView, and Places2 datasets demonstrates that ORFNet outperforms existing representative inpainting methods. It leads to 1.98 dB increase in PSNR and 2.49% improvement in SSIM, along with average 2.4% reduction in LPIPS. Experimental results confirm the effectiveness of the proposed image inpainting method, showcasing superior performance across various receptive field settings and achieving more realistic and natural visual outcome. [ABSTRACT FROM AUTHOR]

Ultrasonic testing is widely used for defect detection in polymer composites owing to advantages such as fast processing speed, simple operation, high reliability, and real-time monitoring. However, defect information in ultrasound images is not easily detectable because of the influence of ultrasound echoes and noise. In this study, a stable three-dimensional deep convolutional autoencoder (3D-DCA) was developed to identify defects in polymer composites. Through 3D convolutional operations, it can synchronously learn the spatiotemporal properties of the data volume. Subsequently, the depth receptive field (RF) of the hidden layer in the autoencoder maps the defect information to the original depth location, thereby mitigating the effects of the defect surface and bottom echoes. In addition, a dual-layer encoder was designed to improve the hidden layer visualization results. Consequently, the size, shape, and depth of the defects can be accurately determined. The feasibility of the method was demonstrated through its application to defect detection in carbon-fiber-reinforced polymers. [ABSTRACT FROM AUTHOR]

IntroductionIn primary auditory cortex (A1), nicotinic acetylcholine receptors (nAChRs) containing α2 subunits are expressed in layer 5 Martinotti cells (MCs)—inhibitory interneurons that send a main axon to superficial layers to inhibit distal apical dendrites of pyramidal cells (PCs). MCs also contact interneurons in supragranular layers that, in turn, inhibit PCs. Thus, MCs may regulate PCs via inhibition and disinhibition, respectively, of distal and proximal apical dendrites. Auditory inputs to PCs include thalamocortical inputs to middle layers relaying information about characteristic frequency (CF) and near-CF stimuli, and intracortical long-distance (“horizontal”) projections to multiple layers carrying information about spectrally distant (“nonCF”) stimuli. CF and nonCF inputs integrate to create broad frequency receptive fields (RFs). Systemic administration of nicotine activates nAChRs to “sharpen” RFs—to increase gain within a narrowed RF—resulting in enhanced responses to CF stimuli and reduced responses to nonCF stimuli. While nicotinic mechanisms to increase gain have been identified, the mechanism underlying RF narrowing is unknown.MethodsHere, we examine the role of α2 nAChRs in mice with α2 nAChR-expressing neurons labeled fluorescently, and in mice with α2 nAChRs genetically deleted.ResultsThe distribution of fluorescent neurons in auditory cortex was consistent with previous studies demonstrating α2 nAChRs in layer 5 MCs, including nonpyramidal somata in layer 5 and dense processes in layer 1. We also observed label in subcortical auditory regions, including processes, but no somata, in the medial geniculate body, and both fibers and somata in the inferior colliculus. Using electrophysiological (current-source density) recordings in α2 nAChR knock-out mice, we found that systemic nicotine failed to enhance CF-evoked inputs to layer 4, suggesting a role for subcortical α2 nAChRs, and failed to reduce nonCF-evoked responses, suggesting that α2 nAChRs regulate horizontal projections to produce RF narrowing.DiscussionThe results support the hypothesis that α2 nAChRs function to simultaneously enhance RF gain and narrow RF breadth in A1. Notably, a similar neural circuit may recur throughout cortex and hippocampus, suggesting widespread conserved functions regulated by α2 nAChRs.

Survey of road cracks in a timely, complete, and accurate way is pivotal to pavement maintenance planning. Motivated by the increasingly heavy task of identifying cracks, researchers have developed extensive crack segmentation models based on Deep learning (DL) methods with significantly different levels of accuracy, efficiency, and generalizing capacity. Although many of the models provide satisfying detection performance, why these models work still needs to be determined. The objective of this study is to survey recent advances in automated DL crack recognition and provide evidence for their underlying working mechanism. We first reviewed 54 DL crack recognition methods to summarize critical factors in these models. Then, we conducted a performance evaluation of fourteen famous semantic segmentation models using the quantitative metrics: F-1 score and mIoU. Then, the effective receptive field and class activation map of the included models are visualized to demonstrate the training results as qualitative evaluation. Based on the literature review and comparison results, larger kernel size, feature fusion, and attention module all contribute to the improvement of model performance. Striking a balance between increasing the effective receptive field and computational/memory efficiency is the key to designing DL crack segmentation models. Finally, some potential directions and suggestions for future development are provided, such as developing semi-supervised or unsupervised learning for the high cost of pixel-level labeling.

Brain tumors are serious conditions, which can cause great trauma to patients, endangering their health and even leading to disability or death. Therefore, accurate preoperative diagnosis is particularly important. Accurate brain tumor segmentation based on deep learning plays an important role in the preoperative treatment planning process and has achieved good performance. However, one of the challenges involved is an insufficient ability to extract features with a large receptive field in encoder layers and guide the selection of deep semantic information in decoder layers. We propose small kernel convolution with an effective receptive field network (SCENet) based on UNet, which involves a small kernel convolution with effective receptive field shuffle module (SCER) and a channel spatial attention module (CSAM). The SCER module utilizes the inherent properties of stacking convolution to obtain effectively receptive fields and improve the features with a large receptive field extraction ability. CSAM of decoder layers can preserve more detailed features to capture clearer contours of the segmented image by calculating the weights of channels and spaces. An ASPP module is introduced to the bottleneck layer to enlarge the receptive field and can capture multi-scale detailed features. Furthermore, a large number of experiments were performed to evaluate the performance of our model on the BraTS2021 dataset. The SCENet achieved dice coefficient scores of 91.67%, 87.70%, and 83.35% for whole tumor (WT), tumor core (TC), and enhancing tumor (ET), respectively. The results show that the proposed model achieves the state-of-the-art performance compared with more than twelve benchmarks.

The Whole Slide Image (WSI) is a pathological image with Hematoxylin & Eosin staining. The low-contrast color staining will bring a challenge on analysis. We propose SNSeg (Staining Nuclear Segmentation) to improve the segmentation performance in WSI, for obtaining accurate nuclear region. At the macro level, we reconstructed the feature fusion mode and connection path, for reducing semantic loss in the gradient descent. At the micro level, first, we design a multiple receptive field convolution unit (RFC), and it can adjust the receptive field for adapting to the nuclei size of the input image. Secondly, for efficiently fusing the feature information extracted from the encoder, we design a multi-branch channel attention fusion unit (MCA), which integrates different branch information flows in channel-wise to a unified module. Finally, we design parallel outputting decoder fusion (DF) module to fuse outputting spatial attention for generating the final segmentation results. In addition, we introduce the watershed based on distance transformation to separate adherent nuclei and mark contours. We design experiments for verifying SNSeg on public datasets of MoNuSeg, TNBC, and PanNuKe. The segmentation results on MoNuSeg show that the SNSeg has achieves an accuracy of 84.32% and a Dice score of 81.21%. Compared with other networks, the SNSeg have competitive advantages in segmentation performance and network parameters. [ABSTRACT FROM AUTHOR]

Based on image contour, the instance segmentation method uses fewer contour nodes to represent an object, which effectively reduces the number of algorithmic parameters and improves its operation efficiency. However, with the segmentation result of poor quality, it is no match for traditional pixel-by-pixel processing segmentation algorithm in terms of accuracy. To improve the accuracy of the algorithm, it is of great necessity to introduce a refined model of the instance segmentation (Attend the Contour snake, AC-snake), which is based on image contour with a combination of attention mechanism. An improved Largekernel + is added to the backbone network to improve the receptive field of the model and extract richer feature information. The network structure at the contour vertex deformation stage is improved, and the Dual Channel attention (DC - attentio) module is combined to enhance the effective information of contour vertex, reduce the invalid parameters in the training network, and improve the detection accuracy and training speed. The experimental results show that in Cityscapes validation data set, the improved model proposed in this study has improved performance when compared with the original model. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Signal Processing is the property of Journal of Signal Processing 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.)

Copyright of Computer Measurement & Control is the property of Magazine Agency of Computer Measurement & Control 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.)

The concept of receptive field (RF) is central to sensory neuroscience. Neuronal RF properties have been substantially studied in animals, while those in humans remain nearly unexplored. Here, we measured neuronal RFs with intracranial local field potentials (LFPs) and spiking activity in human visual cortex (V1/V2/V3). We recorded LFPs via macro-contacts and discovered that RF sizes estimated from low-frequency activity (LFA, 0.5–30 Hz) were larger than those estimated from low-gamma activity (LGA, 30–60 Hz) and high-gamma activity (HGA, 60–150 Hz). We then took a rare opportunity to record LFPs and spiking activity via microwires in V1 simultaneously. We found that RF sizes and temporal profiles measured from LGA and HGA closely matched those from spiking activity. In sum, this study reveals that spiking activity of neurons in human visual cortex could be well approximated by LGA and HGA in RF estimation and temporal profile measurement, implying the pivotal functions of LGA and HGA in early visual information processing. [ABSTRACT FROM AUTHOR]

Neural circuits function in the face of changing inputs, either caused by normal variation in stimuli or by cell death. To maintain their ability to perform essential computations with partial inputs, neural circuits make modifications. Here, we study the retinal circuit's responses to changes in light stimuli or in photoreceptor inputs by inducing partial cone death in the mature mouse retina. Can the retina withstand or recover from input loss? We find that the excitatory pathways exhibit functional loss commensurate with cone death and with some aspects predicted by partial light stimulation. However, inhibitory pathways recover functionally from lost input by increasing spatiotemporal integration in a way that is not recapitulated by partially stimulating the control retina. Anatomically, inhibitory synapses are upregulated on secondary bipolar cells and output ganglion cells. These findings demonstrate the greater capacity for inhibition, compared with excitation, to modify spatiotemporal processing with fewer cone inputs.

Scene text recognition methods of the Encoder–Decoder framework, generally assume that the proportion of characters in the same text instance are basically the same. However, this assumption does not always hold in the context of complex scene images. For adaptively revising the receptive field according to the different font in the scene text image, we propose a Dynamic Receptive Field Adaption Framework which consists of Memory Attention (MA) module and Dynamic Feature Adaptive (DFA) module. MA percepts the historical location information to adapt to the change of character position in the decoder. DFA selects the most distinguishing features from feature maps of different levels dynamically. Additionally, MA and DFA can be easily extended to the existing attention-based and transformer-based text recognition methods to improve their performance. With extension experiments on public benchmark datasets, including IIIT-5K, SVT, SVTP, CUTE80, RECTS, LSVT, and RCTW, our method has shown effectiveness and robustness. [ABSTRACT FROM AUTHOR]