Search

The prediction of power equipment activity plays a vital role in optimizing power resource dispatch, ensuring supply and demand balance, and guiding network planning and management. However, due to the complex nonlinear, multi-scale, and multivariate characteristics of power grid communication networks service activity (PCNSA) data, it is often challenging to capture its intrinsic patterns and dynamic changes through a simple model. To address this issue, this paper proposes a power grid communication network service activity prediction method based on convolutional neural network, bidirectional gated recurrent unit, and attention mechanism, referred to as CBGA. CNN is used to extract features from time-series data, BiGRU is used to capture long-term feature changes in the data, and the attention mechanism is used to enhance the extraction of important information. To validate the performance of the proposed method, experiments and comparisons were conducted on three real power grid communication network datasets, demonstrating that our method shows better performance and robustness. [ABSTRACT FROM AUTHOR]

Accurate power load forecasting is crucial for the sustainable operation of smart grids. However, the complexity and uncertainty of load, along with the large-scale and high-dimensional energy information, present challenges in handling intricate dynamic features and long-term dependencies. This paper proposes a computational approach to address these challenges in short-term power load forecasting and energy information management, with the goal of accurately predicting future load demand. The study introduces a hybrid method that combines multiple deep learning models, the Gated Recurrent Unit (GRU) is employed to capture long-term dependencies in time series data, while the Temporal Convolutional Network (TCN) efficiently learns patterns and features in load data. Additionally, the attention mechanism is incorporated to automatically focus on the input components most relevant to the load prediction task, further enhancing model performance. According to the experimental evaluation conducted on four public datasets, including GEFCom2014, the proposed algorithm outperforms the baseline models on various metrics such as prediction accuracy, efficiency, and stability. Notably, on the GEFCom2014 dataset, FLOP is reduced by over 48.8%, inference time is shortened by more than 46.7%, and MAPE is improved by 39%. The proposed method significantly enhances the reliability, stability, and cost-effectiveness of smart grids, which facilitates risk assessment optimization and operational planning under the context of information management for smart grid systems. [ABSTRACT FROM AUTHOR]

Generally, the recognition performance of lightweight models is often lower than that of large models. Knowledge distillation, by teaching a student model using a teacher model, can further enhance the recognition accuracy of lightweight models. In this paper, we approach knowledge distillation from the perspective of intermediate feature-level knowledge distillation. We combine a cross-stage feature fusion symmetric framework, an attention mechanism to enhance the fused features, and a contrastive loss function for teacher and student models at the same stage to comprehensively implement a multistage feature fusion knowledge distillation method. This approach addresses the problem of significant differences in the intermediate feature distributions between teacher and student models, making it difficult to effectively learn implicit knowledge and thus improving the recognition accuracy of the student model. Compared to existing knowledge distillation methods, our method performs at a superior level. On the CIFAR100 dataset, it boosts the recognition accuracy of ResNet20 from 69.06% to 71.34%, and on the TinyImagenet dataset, it increases the recognition accuracy of ResNet18 from 66.54% to 68.03%, demonstrating the effectiveness and generalizability of our approach. Furthermore, there is room for further optimization of the overall distillation structure and feature extraction methods in this approach, which requires further research and exploration. [ABSTRACT FROM AUTHOR]

The concentration prediction of PM2.5 plays a vital role in controlling the air and improving the environment. This paper proposes a prediction model (namely EEMD-ALSTM) based on Ensemble Empirical Mode Decomposition (EEMD), Attention Mechanism and Long Short-Term Memory network (LSTM). Through the combination of decomposition and LSTM, attention mechanism is introduced to realize the prediction of PM2.5 concentration. The advantage of EEMD-ALSTM model is that it decomposes and combines the original data using the method of ensemble empirical mode decomposition, reduces the high nonlinearity of the original data, and Specially reintroduction the attention mechanism, which enhances the extraction and retention of data features by the model. Through experimental comparison, it was found that the EEMD-ALSTM model reduced its MAE and RMSE by about 15% while maintaining the same R2 correlation coefficient, and the stability of the model in the prediction process was also improved significantly. [ABSTRACT FROM AUTHOR]

In aerial image detection, difficulties in feature extraction and low detection accuracy arise due to background interference, occlusion, and the presence of multiple small objects. This paper proposes a DM-YOLOX aerial object target detection method with intensive attention mechanism. Firstly, the proposed approach incorporates coordinate attention (CA) and a dense connection method into the backbone network architecture, enabling adaptive channel weighting throughout the feature extraction process. This facilitates the enhancement of significant features while suppressing less relevant ones, thereby augmenting the network's capacity to represent object features and ensuring retention and reinforcement of key features. Secondly, the multibranch extraction module (MBE) is incorporated into the feature fusion network to enhance the network's ability in extracting multi-scale feature information from images with extensive coverage, thereby enhancing the detection accuracy and efficiency of small- and medium-sized objects in complex scenes. Finally, the utilization of SIoU instead of IoU as the bounding box loss function effectively addresses the issue of mismatch between real and predicted boxes, leading to accelerated network convergence and improved performance during model training. After training and testing on the VisDrone 2019 dataset, this method effectively detects small objects in complex environments. The DM-YOLOX model shows a significant improvement of 2.7% in mAP compared to the baseline network, while achieving an 8% increase in frames per second (FPS). [ABSTRACT FROM AUTHOR]

Conventional image processing and machine learning based on handcrafted features struggle to meet the real time and high-accuracy requirements for industrial defect detection in complex, sensitive, and dynamic environments. To address this issue, this paper proposes AENet, a novel real-time defect detection network based on an encoder-decoder model, which achieves high detection accuracy and efficiency while demonstrating excellent convergence and generalization. Firstly, a spatial channel attention module in the encoding network is designed to exploit both spatial attention and channel attention using a multi-head 3D self-attention mechanism. This improves parallelism and detection efficiency. Secondly, the decoding network of AENet incorporates the cross-level attention fusion module, which fuses input features from different layers. Combined with multi-level upsampling design, the decoder enhances the representation of defect details. Furthermore, we insert a simplified aggregator into the encoder-decoder network to extract feature information at different scales with low computational cost. This aggregation process aids in training and inference on industrial defect datasets by incorporating contextual information. Extensive experimental results demonstrate that AENet outperforms other segmentation models in accomplishing defect recognition and segmentation in challenging optical environments. It exhibits a faster convergence than other networks and a balance between accuracy and speed. It achieves a recognition accuracy of over 96% for almost all types of defects in the actual industrial environment on the NVIDIA Tesla V100 GPU. [ABSTRACT FROM AUTHOR]

One of the most popular south Indian languages in India is the Telugu language which is currently spoken by 84 million native Telugu speakers in Andhra Pradesh and Telangana. With the rapid growth of the Telugu digital content, the need for the automatic text summarizer is arisen to provide short text from huge text documents. Extractive text summarization model generates only significant sentences. Abstractive text summarization method requires more training time. In this paper, a novel hybrid model is proposed for generating text summaries by combining extractive and abstractive approach to reduce the training time. For extractive method TextRank algorithm is utilized and for abstractive method attention-based sequence to sequence model with bidirectional long short-term memory (Bi-LSTM) is utilized. Moreover, coverage mechanism is included into the proposed hybrid approach to reduce the repetition in summaries and to improve the quality of summaries. The performance of the proposed hybrid model is evaluated by the ROUGE toolkit in terms of F-measure, recall and precision. The results of the proposed model are compared with other existing models which shows that the proposed hybrid model outperforms other existing text summarization models for Telugu Language. [ABSTRACT FROM AUTHOR]

Road damage detection is an crucial task to ensure road safety. To tackle the issues of poor performance on multi-scale pavement distresses and high costs in detection task, this paper presents an improved lightweight road damage detection algorithm based on YOLOv8n, named YOLOv8-PD (pavement distress). Firstly, a BOT module that can extract global information of road damage images is proposed to adapt to the large-span features of crack objects. Secondly, the introduction of the large separable kernel attention (LKSA) mechanism enhances the detection accuracy of the algorithm. Then, a C2fGhost block is constructed in the neck network to strengthen the feature extraction of complex road damages while reducing the computational load. Furthermore, we introduced lightweight shared convolution detection head (LSCD-Head) to improve feature expressiveness and reduce the number of parameters. Finally, extensive experiments on the RDD2022 dataset yield a model with parametric and computational quantities of 2.3M and 6.1 GFLOPs, which are only 74.1% and 74.3% of the baseline, and the mAP reaches an improvement of 1.4 percentage points from the baseline. In addition, experimental results on the RoadDamage dataset show that the mAP increased by 4.2% and this algorithm has good robustness. This method can provide a reference for the automatic detection method of pavement distress. [ABSTRACT FROM AUTHOR]

The application of deep neural network has achieved remarkable success in object detection. However, the network structures should be still evolved consistently and tuned finely to acquire better performance. This gears to the continuous demands on high performance in those complex scenes, where multi-scale objects to be detected are located here and there. To this end, this paper proposes a network structure called Multi-Scale Coupled Attention (MSCA) under the framework of self-attention learning with methodologies of importance assessment. Architecturally, it consists of a Multi-Scale Coupled Channel Attention (MSCCA) module, and a Multi-Scale Coupled Spatial Attention (MSCSA) module. Specifically, the MSCCA module is developed to achieve the goal of self-attention learning linearly on the multi-scale channels. In parallel, the MSCSA module is constructed to achieve this goal nonlinearly on the multi-scale spatial grids. The MSCCA and MSSCA modules can be connected together into a sequence, which can be used as a plugin to develop end-to-end learning models for object detection. Finally, our proposed network is compared on two public datasets with 13 classical or state-of-the-art models, including the Faster R-CNN, Cascade R-CNN, RetinaNet, SSD, PP-YOLO, YOLO v3, YOLO v5, YOLO v7, YOLOX, DETR, conditional DETR, UP-DETR and FP-DETR. Comparative experimental results with numerical scores, the ablation study, and the performance behaviour all demonstrate the effectiveness of our proposed model. [ABSTRACT FROM AUTHOR]

Transportation mode recognition is an important and challenging problem in intelligent transportation systems. For decades, many data mining and deep learning methods have been proposed, but all these methods have some deficiencies and do not fully use temporal and spatial information hidden in raw GPS trajectory data. In this paper, we presented a new deep learning model called TSANET, which designs an attention mechanism to combine global and local spatiotemporal features. The input of the model is seven kinematic features, derived from the GPS trajectories and covers all information hidden in the GPS points. The model adopts TCN and ST-Block to extract global and local spatiotemporal features, respectively. In ST-Block, BiGRU is used to capture temporal features and DenseNet is applied to capture spatial features. In TCN, convolution operations are applied to obtain temporal and spatial information. Moreover, a dual-layer attention mechanism is developed to integrate and reconstruct these features by assigning different weights to global and local spatiotemporal features caught by TCN and ST-Block. By comprehensively considering global and local features, the model can greatly improve classification performance and recognition granularity, and with the further introduction of the attention mechanism, it can identify all transportation modes in a fine-grained manner. At last, the experiments have been conducted on the Geo1 and Geo2 datasets used by many researchers. The experimental results show that our model cannot only achieve the highest accuracy of 94.34 % among all recent methods but also identify all seven transportation modes clearly, thus verifying the advantages and effectiveness of the model. [ABSTRACT FROM AUTHOR]

This paper proposes an improved strategy for the MobileNetV2 neural network(I-MobileNetV2) in response to problems such as large parameter quantities in existing deep convolutional neural networks and the shortcomings of the lightweight neural network MobileNetV2 such as easy loss of feature information, poor real-time performance, and low accuracy rate in facial emotion recognition tasks. The network inherits the characteristics of MobilenetV2 depthwise separated convolution, signifying a reduction in computational load while maintaining a lightweight profile. It utilizes a reverse fusion mechanism to retain negative features, which makes the information less likely to be lost. The SELU activation function is used to replace the RELU6 activation function to avoid gradient vanishing. Meanwhile, to improve the feature recognition capability, the channel attention mechanism (Squeeze-and-Excitation Networks (SE-Net)) is integrated into the MobilenetV2 network. Experiments conducted on the facial expression datasets FER2013 and CK + showed that the proposed network model achieved facial expression recognition accuracies of 68.62% and 95.96%, improving upon the MobileNetV2 model by 0.72% and 6.14% respectively, and the parameter count decreased by 83.8%. These results empirically verify the effectiveness of the improvements made to the network model. [ABSTRACT FROM AUTHOR]

Artificial intelligence has made significant breakthroughs in many fields, especially with the broad deployment of edge devices, which provides opportunities to develop and apply various intelligent models in edge networks. Edge device-server co-inference system has gradually become the mainstream of edge intelligent computing. However, the existing feature procession works in the edge inference framework neglect the focus on whether features are important, and the processed features are still redundant, affecting the inference efficiency. In this paper, we propose a novel attention-based variable-size feature compression module to enhance edge systems' inference efficiency by leveraging input data's varying importance levels. First, a multi-scale attention mechanism is introduced, which operates jointly in the channel spatial to effectively compute importance weights from the intermediate output features of the edge devices. These weights are then utilized to assign different transmission probabilities, filtering out irrelevant feature data and prioritizing task-relevant information. Second, the new loss algorithm and progressive model training strategy are designed to optimize the proposed module, enabling the model to adapt to the reduced feature data gradually and effectively. Finally, experimental results on CIFAR-10 and ImageNet datasets demonstrate the effectiveness of our proposed solution, showcasing a significant reduction in the data output volume of edge devices and minimizing communication overhead while ensuring minimal loss in model accuracy. [ABSTRACT FROM AUTHOR]

Aspect-level sentiment classification is a fine-grained sentiment analysis that primarily focuses on predicting the sentiment polarity of aspects within a sentence. At present, many methods employ graph convolutional networks (GCN) to extract hidden semantic or syntactic information from sentences, achieving good results. However, these existing methods often overlook the relationships between multiple aspects within a sentence, treating aspects separately and thus neglecting the sentiment connections. To address this issue, this paper introduces a triple-channel graph attention network (TC-GAT) to capture semantics, syntax and multiple aspects dependencies information. In addition, a simple and effective fusion mechanism is proposed to comprehensively integrate these three types of information. Experiments are carried out on three commonly datasets, and the results verify the effectiveness of our proposed model. [ABSTRACT FROM AUTHOR]

Wearable sensors are widely used in medical applications and human–computer interaction because of their portability and powerful privacy. Human activity identification based on sensor data plays a vital role in these fields. Therefore, it is important to improve the recognition performance of different types of actions. Aiming at the problems of insufficient time-varying feature extraction and gradient explosion caused by too many network layers, a time convolution network recognition model with attention mechanism (TCN-Attention-HAR) was proposed. The model effectively recognizes and emphasizes the key feature information. The ability of extracting temporal features from TCN (temporal convolution network) is improved by using the appropriate size of the receiver domain. In addition, attention mechanisms are used to assign higher weights to important information, enabling models to learn and identify human activities more effectively. The performance of the Open Data Set (WISDM, PAMAP2 and USC-HAD) is improved by 1.13%, 1.83% and 0.51%, respectively, compared with other advanced models, these results clearly show that the network model presented in this paper has excellent recognition performance. In the knowledge distillation experiment, the parameters of student model are only about 0.1% of those of teacher model, and the accuracy of the model has been greatly improved, and in the WISDM data set, compared with the teacher's model, the accuracy is 0.14% higher. [ABSTRACT FROM AUTHOR]

The presence of haze results in diminished image clarity, the loss of image details, color alterations, and a decline in image visibility. This paper proposes an efficient multi-stage feature fusion defogging network based on the attention mechanism for effective defogging in complex environments. To create a multi-branch defogging network, the model integrates several attention techniques to create various sorts of branching network architectures. The image's spatial details and contextual information are supplemented and merged based on the recovered image feature information of different network branches to increase the effectiveness of the network model. The stage attention fusion mechanism created among several network branches can lessen the loss of image data during feature extraction and improve the effectiveness of the image-defogging operation. The experimental results demonstrate that the proposed algorithm has superior defogging performance in both synthetic and real-world scene datasets and performs more admirably in terms of accuracy compared to other sophisticated algorithms, particularly in the RESIDE and O-Haze datasets. The PSNR metrics on the RESIDE and O-Haze datasets are improved by 1.58 dB and 1.61 dB, respectively, compared to the best-advanced technique suggested in this study, and the SSIM metrics on the O-Haze dataset are improved by 3.4%. [ABSTRACT FROM AUTHOR]

Aiming at the problems of identifying storage pest tobacco pest beetles from images that have few object pixels and considerable image noise, and therefore suffer from lack of information and identifiable features, this paper proposes an automatic monitoring method of tobacco beetle based on Multi-scale Global residual Feature Pyramid Network and Dual-path Deformable Attention (MGrFPN-DDrGAM). Firstly, a Multi-scale Global residual Feature Pyramid Network (MGrFPN) is constructed to obtain rich high-level semantic features and more complete information on low-level features to reduce missed detection; Then, a Dual-path Deformable receptive field Guided Attention Module (DDrGAM) is designed to establish long-range channel dependence, guide the effective fusion of features and improve the localization accuracy of tobacco beetles by fitting the spatial geometric deformation features of and capturing the spatial information of feature maps with different scales to enrich the feature information in the channel and spatial. Finally, to simulate a real scene, a multi-scene tobacco beetle dataset is created. The dataset includes 28,080 images and manually labeled tobacco beetle objects. The experimental results show that under the framework of the Faster R-CNN algorithm, the detection precision and recall rate of this method can reach 91.4% and 98.4% when the intersection ratio (IoU) is 0.5. Compared with Faster R-CNN and FPN, when the intersection ratio (IoU) is 0.7, the detection precision is improved by 32.9% and 6.9%, respectively. The proposed method is superior to the current mainstream methods. [ABSTRACT FROM AUTHOR]

To address the fuzzy segmentation boundaries, missing details, small target losses and low efficiency of traditional segmentation methods in ancient mural image segmentation scenarios, this paper proposes a mural segmentation model based on multiscale feature fusion and a dual attention-augmented segmentation model (MFAM). The model uses the MobileViT network, which integrates a coordinate attention mechanism, as the feature extraction backbone network. It attains global and local expression capabilities through self-attention, class convolution, and coordinate attention and focuses on location information to expand the receptive field and achieve improved feature extraction efficiency. An A_R_ASPP feature enhancement module is proposed for the attention-optimized residual atrous spatial pyramid pooling module. The module uses residual connections to solve the small target loss problem in murals caused by the excessive sampling rate of atrous convolution and uses a feature attention mechanism to adaptively adjust the feature map weight according to the channel importance levels. A dual attention-enhanced feature fusion module is proposed for multiscale decoder feature fusion to improve the mural segmentation effect. This module uses a cross-level aggregation strategy and an attention mechanism to weight the importance of different feature levels to obtain multilevel semantic feature representations. The model improves the mean intersection over union (MIoU) by 3.06% and the MPA by 1.81% on a mural dataset compared with other models. The model is proven to be effective at improving the segmentation details, efficiency and small target segmentation results produced for mural images, and a new method is proposed for segmenting ancient mural images. [ABSTRACT FROM AUTHOR]

Semantic relatedness is one of the most significant aspects of natural language processing. It has been identified as a critical technology for developing intelligent systems like Siri, Microsoft Ice, Cortana, and Xiaoai. In 2014, SemEval ranked SR as the top task. While many existing studies have focused on analyzing the entailment of single phrases, advancements in deep learning have made it possible to analyze complete sentences or texts. While the natural parallelism of membrane computing has shown promise for data processing, harnessing this potential to advance semantic relatedness remains an open problem yet to be tackled. This paper proposes a novel Siamese Manhattan LSTM-SNP approach (SiMaLSTM-SNP) for the SR problem. The approach uses a collaborative Word2vec and 10-Layer Attention strategy to represent and extract sentence pairs and a Siamese LSTM-SNP structure to calculate the hidden states of sentences. The multi-head self-attention layer identifies text associations and redistributes hidden state weights. The last hidden state is extracted, and the relatedness score is calculated using the Manhattan distance. The experiments demonstrate that SiMaLSTM-SNP outperforms 17 classical SR baselines and 7 novel approaches on the standard datasets SICK and STS in terms of mean square error performance. This indicates that SiMaLSTM-SNP can accurately capture the semantic distinction between two sentences and effectively preserve their semantic information. [ABSTRACT FROM AUTHOR]

The transformer-based encoder-decoder framework is becoming popular in scene text recognition, largely because it naturally integrates recognition clues from both visual and semantic domains. However, recent studies show that the two kinds of clues are not always well registered and therefore, feature and character might be misaligned in difficult text (e.g., with a rare shape). As a result, constraints such as character position are introduced to alleviate this problem. Despite certain success, visual and semantic are still separately modeled and they are merely loosely associated. In this paper, we propose a novel module called multi-domain character distance perception (MDCDP) to establish a visually and semantically related position embedding. MDCDP uses the position embedding to query both visual and semantic features following the cross-attention mechanism. The two kinds of clues are fused into the position branch, generating a content-aware embedding that well perceives character spacing and orientation variants, character semantic affinities, and clues tying the two kinds of information. They are summarized as the multi-domain character distance. We develop CDistNet that stacks multiple MDCDPs to guide a gradually precise distance modeling. Thus, the feature-character alignment is well build even though various recognition difficulties are presented. We verify CDistNet on ten challenging public datasets and two series of augmented datasets created by ourselves. The experiments demonstrate that CDistNet performs highly competitively. It not only ranks top-tier in standard benchmarks, but also outperforms recent popular methods by obvious margins on real and augmented datasets presenting severe text deformation, poor linguistic support, and rare character layouts. In addition, the visualization shows that CDistNet achieves proper information utilization in both visual and semantic domains. Our code is available at https://github.com/simplify23/CDistNet. [ABSTRACT FROM AUTHOR]

Multi-agent cooperation and confrontation technology have achieved rapid development in recent years. Most extant multi-agent reinforcement learning algorithms simplify the problem by using shared weights or local observation, and are only suitable for scenarios with less than ten agents. Given this, large-scale scene research needs to explore new directions. This paper presents a large-scale multi-agent evolutionary reinforcement jointed method. The multi-agent learning task is separated into numerous stages based on the agent's scale, and the self-attention mechanism is utilized to handle changing numbers of agents in each step. Simultaneously, to avoid the agents' poor adaptability in previous stages, the best individuals in the population are chosen at each stage of training via evolutionary techniques. Two typical unmanned aerial vehicle cluster missions, multi-domain joint sea crossing and landing missions, were created to validate the performance of the suggested technique, and the operational rules and reward functions were also given. Experiments have shown that the model trained using the suggested method has good performance and stability and can provide a multi-agent collaborative decision-making model suitable for large-scale environments. [ABSTRACT FROM AUTHOR]

Parameter identification of transmission lines plays a crucial role in power systems, and many deep learning methods have been continuously applied to this domain. However, these methods are highly sensitive to data corruption, and as the scale of the power grid continues to expand, the model’s solving accuracy deteriorates. In response to these challenges, this paper introduces a multi-scale graph attention network (UMGAN) that leverages the spatial structural characteristics of the power grid. In order to enable the model to fully exploit the correlations between power grid branches and learn more precise node feature representations, we have designed a simplified graph neural network convolution kernel, creating a lighter U-shaped structure for multi-scale sampling of power grid data. Additionally, an attention layer has been incorporated at skip connections to reduce the impact of data corruption. Furthermore, for the multi-task learning module, we have devised a multi-task loss function tailored for power grid parameter identification tasks. This loss function effectively balances multiple objectives, allowing simultaneous identification of multiple parameters. Experimental results demonstrate that the multi-scale graph attention network proposed in this paper outperforms other machine learning and deep learning methods, providing more accurate predictions of power grid branch parameters. [ABSTRACT FROM AUTHOR]

Automated testing of GUI elements on electronic screens by machine vision is widely used in smart manufacturing production lines. However, in the complex environment of factories, interference factors such as light reflection, platform vibration, and placement angle make it difficult for testing robots to recognize and locate GUI elements on the equipment’s screen to be tested. In addition, the recognition algorithms based on screen GUI elements currently have problems such as slow detection speed and large model size, which limit the deployment of detection models on test robots. To address these problems, this paper proposes a lightweight model that can overcome the interference of the factory environment based on the YOLOv5 algorithm for recognizing GUI elements. First, a reinforced position and channel attention module (PCA) is proposed to enhance the extraction and fusion ability of weak feature elements affected by interference; then, a three-way feature pyramid network (TWFPN) module is constructed at the intermediate feature layer of the neck structure to provide more accurate localization information. Meanwhile, the loss function of the original algorithm is replaced with SIoU to improve the convergence speed. Finally, a lightweight structure combining the Ghost module and PConv is introduced into the model to reduce the model parameters and improve the detection speed. The method was evaluated on the constructed feature-enhanced GUI (FE_GUI) dataset. The final model shows a 2.7% improvement in average accuracy (mAP@0.5), a 30% reduction in model weight, and an increase in detection speed to 85 FPS compared to the original YOLOv5. The results demonstrate that the method solves the issues of element recognition and model deployment for a vision test system and surpasses existing models in terms of recognition accuracy and detection speed. [ABSTRACT FROM AUTHOR]

Floor plan is an important communication tool between architects, construction engineers, and clients for a building project. Estimation of building features from a floor plan image is often a time-consuming task. Automatic analysis of floor plan images can significantly improve work efficiency and accuracy. A few research works have been reported in the literature on automated floor image analysis. However, the scope and performance of the existing techniques are limited. In this paper, a CNN-based technique, referred to as FloorNet, is proposed for the multiclass semantic segmentation of a floor plan. The proposed FloorNet has five modules: Encoder, Room type decoder, Room boundary decoder, Multiscale room boundary attention model and Floor classification. The proposed technique is evaluated using simple brochure type and complex architectural type floor plan images. Experimental results show that the proposed technique provides an improvement of 5–11% mIoU for semantic segmentation (for 9–11 classes) compared to the state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Graph neural networks (GNNs) have indeed shown significant potential in the field of personalized recommendation. The core approach is to reorganize interaction data into a user–item bipartite graph, leveraging high-order connectivity between user and item nodes to enhance their representations. However, most existing methods only deploy graph neural networks on parallel interaction graphs and treat the information propagated from all neighbors as equivalent, failing to adaptively capture user preferences. Therefore, the representations obtained may contain redundant or even noisy information, leading to non-robustness and suboptimal performance. The recently proposed Multimodal Graph Attention Network (MGAT) disentangles personal interests at the granularity of modality, operating on individual modal interaction graphs while utilizing a gated attention mechanism to differentiate the impacts of different modalities on user preferences. However, MGAT merely uses averaging to fuse multimodal features, which might overlook unique or critical information within each modality. To address this issue, this paper proposes a multimodal preference-based graph attention network. Firstly, for each individual modality, a single-modality graph network is constructed by integrating the user–item interaction bipartite graph, enabling it to learn user preferences for that modality. GNNs are used to aggregate neighborhood information and enhance the representation of each node. Additionally, a GRU module is utilized to determine whether to aggregate neighborhood information, thereby achieving noise reduction. In addition, a lightweight complementary attention mechanism is proposed to fuse user and item representations learned from different modal graphs. The complementary attention mechanism can not only avoid information redundancy, but also alleviate the problem of modal loss to a certain extent, and ultimately input the fusion results into the prediction module. Experimental results on the MovieLens and TikTok datasets demonstrate the effectiveness of the multimodal information and attention fusion mechanism in improving recommendation accuracy. Compared to baseline state-of-the-art algorithms, the proposed model achieves significant improvements in the Precision@K, Recall@K, NDCG@K and AUC metrics. Our code is publicly available on GitHub: [https://github.com/Oasisway624/mgpat]. [ABSTRACT FROM AUTHOR]

The attention mechanism has been proven effective on various visual tasks in recent years. In the semantic segmentation task, the attention mechanism is applied in various methods, including the case of both convolution neural networks and vision transformer as backbones. However, we observe that the attention mechanism is vulnerable to patch-based adversarial attacks. Through the analysis of the effective receptive field, we attribute it to the fact that the wide receptive field brought by global attention may lead to the spread of the adversarial patch. To address this issue, in this paper, we propose a robust attention mechanism (RAM) to improve the robustness of the semantic segmentation model, which can notably relieve the vulnerability against patch-based attacks. Compared to the vallina attention mechanism, RAM introduces two novel modules called max attention suppression and random attention dropout, both of which aim to refine the attention matrix and limit the influence of a single adversarial patch on the semantic segmentation results of other positions. Extensive experiments demonstrate the effectiveness of our RAM to improve the robustness of semantic segmentation models against various patch-based attack methods under different attack settings. [ABSTRACT FROM AUTHOR]

Monocular depth estimation (MDE) aims to predict pixel-level dense depth maps from a single RGB image. Some recent approaches mainly rely on encoder–decoder architectures to capture and process multi-scale features. However, they usually exploit heavier network at the expense of computational costs to obtain high-quality depth maps. In this paper, we propose a novel enriched multi-path vision transformer feature interaction network with an encoder–decoder architecture, denoted as EDFIDepth , which seeks a balance between computational costs and performance rather than pursuing the highest accuracy or extremely lightweight models. Specifically, an encoder called MPViT-D, incorporating multi-path vision transformer and a deep convolution module, is introduced to extract diverse features with both fine and coarse details at the same feature level with fewer parameters. Subsequently, we propose a lightweight decoder comprising two effective modules to establish multi-scale feature interaction: an encoder–decoder cross-feature matching (ED-CFM) module and a channel-level feature fusion (CLFF) module. The ED-CFM module is to establish connections between encoder–decoder features through a dual-path structure, where a cross-attention mechanism is deployed to enhance the relevance of multi-scale complementary depth information. Meanwhile, the CLFF module utilizes a channel attention mechanism to further fuse crucial depth information within the channels, thereby improving the accuracy of depth estimation. Extensive experiments on the indoor dataset NYUv2 and the outdoor dataset KITTI demonstrate that our method can achieve comparable state-of-the-art (SOTA) results while significantly reducing the number of trainable parameters. Our codes and approach are available at https://github.com/Zhangmg123/EDFIDEpth. [ABSTRACT FROM AUTHOR]

Ultra-high resolution image segmentation has raised increasing interests in recent years due to its realistic applications. In this paper, we innovate the widely used high-resolution image segmentation pipeline, in which an ultra-high resolution image is partitioned into regular patches for local segmentation and then the local results are merged into a high-resolution semantic mask. In particular, we introduce a novel locality-aware context fusion based segmentation model to process local patches, where the relevance between local patch and its various contexts are jointly and complementarily utilized to handle the semantic regions with large variations. Additionally, we present the alternating local enhancement module that restricts the negative impact of redundant information introduced from the contexts, and thus is endowed with the ability of fixing the locality-aware features to produce refined results. Furthermore, in comprehensive experiments, we demonstrate that our model outperforms other state-of-the-art methods in public benchmarks and verify the effectiveness of the proposed modules. Our released codes will be available at: https://github.com/liqiokkk/FCtL. [ABSTRACT FROM AUTHOR]