Frontiers in Neurorobotics最新文献

U-Net and its variants have been widely used in the field of image segmentation. In this paper, a lightweight multi-scale Ghost U-Net (MSGU-Net) network architecture is proposed. This can efficiently and quickly process image segmentation tasks while generating high-quality object masks for each object. The pyramid structure (SPP-Inception) module and ghost module are seamlessly integrated in a lightweight manner. Equipped with an efficient local attention (ELA) mechanism and an attention gate mechanism, they are designed to accurately identify the region of interest (ROI). The SPP-Inception module and ghost module work in tandem to effectively merge multi-scale information derived from low-level features, high-level features, and decoder masks at each stage. Comparative experiments were conducted between the proposed MSGU-Net and state-of-the-art networks on the ISIC2017 and ISIC2018 datasets. In short, compared to the baseline U-Net, our model achieves superior segmentation performance while reducing parameter and computation costs by 96.08 and 92.59%, respectively. Moreover, MSGU-Net can serve as a lightweight deep neural network suitable for deployment across a range of intelligent devices and mobile platforms, offering considerable potential for widespread adoption.

Introduction: Space optimization in architectural planning is a crucial task for maximizing functionality and improving user experience in built environments. Traditional approaches often rely on manual planning or supervised learning techniques, which can be limited by the availability of labeled data and may not adapt well to complex spatial requirements.

Methods: To address these limitations, this paper presents a novel architectural planning robot driven by unsupervised learning for automatic space optimization. The proposed framework integrates spatial attention, clustering, and state refinement mechanisms to autonomously learn and optimize spatial configurations without the need for labeled training data. The spatial attention mechanism focuses the model on key areas within the architectural space, clustering identifies functional zones, and state refinement iteratively improves the spatial layout by adjusting based on learned patterns. Experiments conducted on multiple 3D datasets demonstrate the effectiveness of the proposed approach in achieving optimized space layouts with reduced computational requirements.

Results and discussion: The results show significant improvements in layout efficiency and processing time compared to traditional methods, indicating the potential for real-world applications in automated architectural planning and dynamic space management. This work contributes to the field by providing a scalable solution for architectural space optimization that adapts to diverse spatial requirements through unsupervised learning.

Introduction: In recent years, with advancements in wearable devices and biosignal analysis technologies, sports performance analysis has become an increasingly popular research field, particularly due to the growing demand for real-time monitoring of athletes' conditions in sports training and competitive events. Traditional methods of sports performance analysis typically rely on video data or sensor data for motion recognition. However, unimodal data often fails to fully capture the neural state of athletes, leading to limitations in accuracy and real-time performance when dealing with complex movement patterns. Moreover, these methods struggle with multimodal data fusion, making it difficult to fully leverage the deep information from electroencephalogram (EEG) signals.

Methods: To address these challenges, this paper proposes a "Cerebral Transformer" model based on EEG signals and video data. By employing an adaptive attention mechanism and cross-modal fusion, the model effectively combines EEG signals and video streams to achieve precise recognition and analysis of athletes' movements. The model's effectiveness was validated through experiments on four datasets: SEED, DEAP, eSports Sensors, and MODA. The results show that the proposed model outperforms existing mainstream methods in terms of accuracy, recall, and F1 score, while also demonstrating high computational efficiency.

Results and discussion: The significance of this study lies in providing a more comprehensive and efficient solution for sports performance analysis. Through cross-modal data fusion, it not only improves the accuracy of complex movement recognition but also provides technical support for monitoring athletes' neural states, offering important applications in sports training and medical rehabilitation.

Introduction: RGB-T Salient Object Detection (SOD) aims to accurately segment salient regions in both visible light and thermal infrared images. However, many existing methods overlook the critical complementarity between these modalities, which can enhance detection accuracy.

Methods: We propose the Edge-Guided Feature Fusion Network (EGFF-Net), which consists of cross-modal feature extraction, edge-guided feature fusion, and salience map prediction. Firstly, the cross-modal feature extraction module captures and aggregates united and intersecting information in each local region of RGB and thermal images. Then, the edge-guided feature fusion module enhances the edge features of salient regions, considering that edge information is very helpful in refining significant area details. Moreover, a layer-by-layer decoding structure integrates multi-level features and generates the prediction of salience maps.

Results: We conduct extensive experiments on three benchmark datasets and compare EGFF-Net with state-of-the-art methods. Our approach achieves superior performance, demonstrating the effectiveness of the proposed modules in improving both detection accuracy and boundary refinement.

Discussion: The results highlight the importance of integrating cross-modal information and edge-guided fusion in RGB-T SOD. Our method outperforms existing techniques and provides a robust framework for future developments in multi-modal saliency detection.

Introduction: Segmentation tasks in computer vision play a crucial role in various applications, ranging from object detection to medical imaging and cultural heritage preservation. Traditional approaches, including convolutional neural networks (CNNs) and standard transformer-based models, have achieved significant success; however, they often face challenges in capturing fine-grained details and maintaining efficiency across diverse datasets. These methods struggle with balancing precision and computational efficiency, especially when dealing with complex patterns and high-resolution images.

Methods: To address these limitations, we propose a novel segmentation model that integrates a hierarchical vision transformer backbone with multi-scale self-attention, cascaded attention decoding, and diffusion-based robustness enhancement. Our approach aims to capture both local details and global contexts effectively while maintaining lower computational overhead.

Results and discussion: Experiments conducted on four diverse datasets, including Ancient Architecture, MS COCO, Cityscapes, and ScanNet, demonstrate that our model outperforms state-of-the-art methods in accuracy, recall, and computational efficiency. The results highlight the model's ability to generalize well across different tasks and provide robust segmentation, even in challenging scenarios. Our work paves the way for more efficient and precise segmentation techniques, making it valuable for applications where both detail and speed are critical.

Non-invasive brain-computer interfaces (BCI) hold great promise in the field of neurorehabilitation. They are easy to use and do not require surgery, particularly in the area of motor imagery electroencephalography (EEG). However, motor imagery EEG signals often have a low signal-to-noise ratio and limited spatial and temporal resolution. Traditional deep neural networks typically only focus on the spatial and temporal features of EEG, resulting in relatively low decoding and accuracy rates for motor imagery tasks. To address these challenges, this paper proposes a 3D Convolutional Neural Network (P-3DCNN) decoding method that jointly learns spatial-frequency feature maps from the frequency and spatial domains of the EEG signals. First, the Welch method is used to calculate the frequency band power spectrum of the EEG, and a 2D matrix representing the spatial topology distribution of the electrodes is constructed. These spatial-frequency representations are then generated through cubic interpolation of the temporal EEG data. Next, the paper designs a 3DCNN network with 1D and 2D convolutional layers in series to optimize the convolutional kernel parameters and effectively learn the spatial-frequency features of the EEG. Batch normalization and dropout are also applied to improve the training speed and classification performance of the network. Finally, through experiments, the proposed method is compared to various classic machine learning and deep learning techniques. The results show an average decoding accuracy rate of 86.69%, surpassing other advanced networks. This demonstrates the effectiveness of our approach in decoding motor imagery EEG and offers valuable insights for the development of BCI.

The user perception of mobile game is crucial for improving user experience and thus enhancing game profitability. The sparse data captured in the game can lead to sporadic performance of the model. This paper proposes a new method, the balanced graph factorization machine (BGFM), based on existing algorithms, considering the data imbalance and important high-dimensional features. The data categories are first balanced by Borderline-SMOTE oversampling, and then features are represented naturally in a graph-structured way. The highlight is that the BGFM contains interaction mechanisms for aggregating beneficial features. The results are represented as edges in the graph. Next, BGFM combines factorization machine (FM) and graph neural network strategies to concatenate any sequential feature interactions of features in the graph with an attention mechanism that assigns inter-feature weights. Experiments were conducted on the collected game perception dataset. The performance of proposed BGFM was compared with eight state-of-the-art models, significantly surpassing all of them by AUC, precision, recall, and F-measure indices.

Introduction: Recognizing human actions is crucial for allowing machines to understand and recognize human behavior, with applications spanning video based surveillance systems, human-robot collaboration, sports analysis systems, and entertainment. The immense diversity in human movement and appearance poses a significant challenge in this field, especially when dealing with drone-recorded (RGB) videos. Factors such as dynamic backgrounds, motion blur, occlusions, varying video capture angles, and exposure issues greatly complicate recognition tasks.

Methods: In this study, we suggest a method that addresses these challenges in RGB videos captured by drones. Our approach begins by segmenting the video into individual frames, followed by preprocessing steps applied to these RGB frames. The preprocessing aims to reduce computational costs, optimize image quality, and enhance foreground objects while removing the background.

Result: This results in improved visibility of foreground objects while eliminating background noise. Next, we employ the YOLOv9 detection algorithm to identify human bodies within the images. From the grayscale silhouette, we extract the human skeleton and identify 15 important locations, such as the head, neck, shoulders (left and right), elbows, wrists, hips, knees, ankles, and hips (left and right), and belly button. By using all these points, we extract specific positions, angular and distance relationships between them, as well as 3D point clouds and fiducial points. Subsequently, we optimize this data using the kernel discriminant analysis (KDA) optimizer, followed by classification using a deep neural network (CNN). To validate our system, we conducted experiments on three benchmark datasets: UAV-Human, UCF, and Drone-Action.

Discussion: On these datasets, our suggested model produced corresponding action recognition accuracies of 0.68, 0.75, and 0.83.

Introduction: Myoelectric control systems translate different patterns of electromyographic (EMG) signals into the control commands of diverse human-machine interfaces via hand gesture recognition, enabling intuitive control of prosthesis and immersive interactions in the metaverse. The effect of arm position is a confounding factor leading to the variability of EMG characteristics. Developing a model with its characteristics and performance invariant across postures, could largely promote the translation of myoelectric control into real world practice.

Methods: Here we propose a self-calibrating random forest (RF) model which can (1) be pre-trained on data from many users, then one-shot calibrated on a new user and (2) self-calibrate in an unsupervised and autonomous way to adapt to varying arm positions.

Results: Analyses on data from 86 participants (66 for pre-training and 20 in real-time evaluation experiments) demonstrate the high generalisability of the proposed RF architecture to varying arm positions.

Discussion: Our work promotes the use of simple, explainable, efficient and parallelisable model for posture-invariant myoelectric control.

As robots become integral to various sectors, improving human-robot collaboration is crucial, particularly in anticipating human actions to enhance safety and efficiency. Electroencephalographic (EEG) signals offer a promising solution, as they can detect brain activity preceding movement by over a second, enabling predictive capabilities in robots. This study explores how EEG can be used for action anticipation in human-robot interaction (HRI), leveraging its high temporal resolution and modern deep learning techniques. We evaluated multiple Deep Learning classification models on a motor imagery (MI) dataset, achieving up to 80.90% accuracy. These results were further validated in a pilot experiment, where actions were accurately predicted several hundred milliseconds before execution. This research demonstrates the potential of combining EEG with deep learning to enhance real-time collaborative tasks, paving the way for safer and more efficient human-robot interactions.