Jigang Tong

From the drone’s perspective, object detection faces challenges such as redundant parameters in the backbone network, small-sized objects, dense targets, and difficulties in object localization. To address these problems, a lightweight enhanced detection transformer is proposed. First, a faster, simple parameter-free feature extraction module is proposed in the backbone network to reduce parameter redundancy and efficiently capture shallow-level features. Second, a multi-scale dilated attention module that integrates with receptive field attention convolution operations is proposed in the neck network to mitigate information loss caused by repeated down-sampling. In addition, to improve the detection rate in dense environments, a high-frequency and low-frequency attention mechanism is introduced into the attention-based intra-scale feature interaction module. Finally, to minimize the positioning deviation, a bounding box loss function that fuses the normalized Wasserstein distance and the minimum point distance based on Intersection over Union is introduced. Compared with the baseline model, the proposed architecture reduces parameters by 14.54% while improving mAP50 by 3% on the VisDrone dataset and increasing mAP50 by 2.6% on the HIT-UAV dataset, demonstrating enhanced detection accuracy for unmanned aerial vehicle-based small-object scenarios.

Following the continuous improvement of 3D sensing technologies, Three-dimensional object detection is extensively utilized in autonomous driving scenarios. Current methods still face challenges in terms of high-precision detection. Therefore, we propose a voxel-based 3D object detection algorithm. Traditional voxelization results in the loss of critical information, so we introduce dynamic attention voxel feature encoding to preserve complete spatial information and focus on foreground features. We propose drop sparse convolution to suppress submanifold dilation in 3D feature extraction. To refine object localization and classification tasks, we also design an auxiliary RoI pooling method to extract 2D RoI features from obtained BEV map and integrate them with 3D RoI features. The quantitative evaluation on the KITTI dataset indicates that our method outperforms several advanced methods in terms of vehicle detection accuracy and also demonstrates good performance on other performance metrics. In addition, the average accuracy has been significantly improved compared to the baseline.

Convolutional neural network (CNN) models, such as U-Net, V-Net, and DeepLab, have achieved remarkable results across various medical imaging modalities, a

In real-world complex scenarios, challenges are faced by object detection due to factors such as significant variations in target scale, similarity with the background, dense and overlapping instances, and small-sized targets. To address these challenges, we optimized the detector head part through the employment of decoupled detector head and auxiliary detector head algorithms. The classification and localization tasks are partitioned by these two algorithms, and the sample quality is enhanced by the auxiliary detector head to improve accuracy. To solve the information loss of small targets in the model downsampling, the residual structure and loss function are optimized so that the feature information of small targets is retained. Moreover, a multi-scale path aggregation network and a multi-scale detection head are adopted to retain local feature information while focusing on global features. Furthermore, dense scenes and small targets are adapted to by the integration of the anchor-based algorithm and the introduction of a loss algorithm, and more accurate localization is facilitated. Experimental results show a 1.7% increase in mAP on the MS COCO dataset, indicating improved performance in complex real-life scenes. A 7.2% enhancement on the TCOD dataset is shown, indicating better performance in densely clustered or overlapping scenes and small target detection. Challenges such as variations in target size, background similarity, dense and overlapping instances, and small-sized targets are tackled, and the accuracy and robustness of object detection are improved. The practical applications of the proposed approach are expanded.

In this paper, the dynamic model of the quadrotor with an elastic hangded payload is analysised, and a new control strategy based on the event triggering mechanism, sliding mode control and energy analysis method is designed for the dynamic model. The energy based control law is employed for the accurate positioning task near the equilibrium point, and sliding mode control law is used to avoid the swing of the hanging payload during the transporting process. The former is designed based on the robust sliding mode control, and the latter is designed based on the energy analysis method, and the stability analysis of the designed system and the control law is carried out using the Lyapunov method and the Russell’s invariant theorem. A group of simulation results is presented to illustrate the effectiveness of the algorithms.

Decoding semantic concepts for imagination and perception tasks (SCIP) is important for rehabilitation medicine as well as cognitive neuroscience. Electroencephalogram (EEG) is commonly used in the relevant fields, because it is a low-cost noninvasive technique with high temporal resolution. However, as EEG signals contain a high noise level resulting in a low signal-to-noise ratio, it makes decoding EEG-based semantic concepts for imagination and perception tasks (SCIP-EEG) challenging. Currently, neural network algorithms such as CNN, RNN, and LSTM have almost reached their limits in EEG signal decoding due to their own short-comings. The emergence of transformer methods has improved the classification performance of neural networks for EEG signals. However, the transformer model has a large parameter set and high complexity, which is not conducive to the application of BCI. EEG signals have high spatial correlation. The relationship between signals from different electrodes is more complex. Capsule neural networks can effectively model the spatial relationship between electrodes through vector representation and a dynamic routing mechanism. Therefore, it achieves more accurate feature extraction and classification. This paper proposes a spatio-temporal capsule network with a self-correlation routing mechaninsm for the classification of semantic conceptual EEG signals. By improving the feature extraction and routing mechanism, the model is able to more effectively capture the highly variable spatio-temporal features from EEG signals and establish connections between capsules, thereby enhancing classification accuracy and model efficiency. The performance of the proposed model was validated using the publicly accessible semantic concept dataset for imagined and perceived tasks from Bath University. Our model achieved average accuracies of 94.9%, 93.3%, and 78.4% in the three sensory modalities (pictorial, orthographic, and audio), respectively. The overall average accuracy across the three sensory modalities is 88.9%. Compared to existing advanced algorithms, the proposed model achieved state-of-the-art performance, significantly improving classification accuracy. Additionally, the proposed model is more stable and efficient, making it a better decoding solution for SCIP-EEG decoding.

Recently, Transformer has achieved great success in computer vision. However, it is constrained because the spatial and temporal complexity grows quadratically with the number of large points in 3D object detection applications. Previous point-wise methods are suffering from time consumption and limited receptive fields to capture information among points. In this paper, we propose a two-stage hyperbolic cosine transformer (ChTR3D) for 3D object detection from LiDAR point clouds. The proposed ChTR3D refines proposals by applying cosh-attention in linear computation complexity to encode rich contextual relationships among points. The cosh-attention module reduces the space and time complexity of the attention operation. The traditional softmax operation is replaced by non-negative ReLU activation and hyperbolic-cosine-based operator with re-weighting mechanism. Extensive experiments on the widely used KITTI dataset demonstrate that, compared with vanilla attention, the cosh-attention significantly improves the inference speed with competitive performance. Experiment results show that, among two-stage state-of-the-art methods using point-level features, the proposed ChTR3D is the fastest one.