As an important part of intelligent surveillance systems, person re-identification (PReID) has drawn wide attention of the public in recent years. Many recent deep learning-based PReID methods have used attention or multi-scale feature learning modules to enhance the discrimination of the learned deep features. However, the attention mechanisms may lose some important feature information. Moreover, the multi-scale models usually embed the multi-scale feature learning module into the backbone network, which increases the complexity of testing network. To address the two issues, we propose a multi-scale deep supervision with attention feature learning deep model for PReID. Specifically, we introduce a reverse attention module to remedy the feature information losing issue caused by the attention module, and a multi-scale feature learning layer with deep supervision to train the network. The proposed modules are only used at the training phase and discarded during the test phase. Experiments on Market-1501, DukeMTMC-reID, CUHK03 and MSMT17 datasets. demonstrate that our model notably beats other competitive state-of-the-art models.

Person re-identification (PReID) has received increasing attention due to its significantly importance in intelligent video surveillance. However, most existing multi-scale feature learning methods embed the multi-scale feature extraction modules for PReID, which increases the complexity of inference network and reduces the timeliness. Moreover, jointly using the small-scale and large-scale features to learn feature representations may weaken the local detailed features extraction and spatial information learning. Besides, some attentive local features are often suppressed when introducing the attention mechanisms for deep PReID models. To address these issues, a deep model with Multi-scale Feature Representation Learning (MFRL) and Random Batch Feature Mask (RBFM) is proposed for PReID in this study. To ensure the feature representations discriminability and spatial information learning, two identity losses are adopted to supervise the small-scale and large-scale features learning in MFRL module, respectively. To alleviate the situation of local attentive features being suppressed by using attention mechanisms, RBFM branch with random feature block dropping strategy which can learn the attentive local feature representations. The proposed methods are only performed in the training phase and discarded in the testing phase, thus, enhancing the effectiveness of the model. Our model achieves the state-of-the-art on the popular benchmark datasets including Market-1501, DukeMTMC-reID and CUHK03. Besides, we conduct a set of ablation experiments to verify the effectiveness of the proposed methods.

Graph Neural Networks (GNNs) are extensions of deep neural networks to graph-structured data. It has already attracted widespread attention for various tasks such as node classification and link prediction. Existing research focuses more on Graph Convolutional Neural Networks (GCNs). However, it is usually overlooked that graph pooling can obtain graph representations by summarizing and down-sampling node information. Meanwhile, existing graph pooling methods mainly use top-k for node selection, but most of them consider only single-view information when scoring nodes, and the k values in top-k are usually selected empirically. This work proposes the Hierarchical Multi-view Top-k Pooling with Deep Q-networks (HMTPool), which scores nodes taking into account multi-view information (Considering graph structure and features) and does not rely on the empirical adaptive selection of the best k-value. HMTPool is a two-stage process. It first uses a variant GCN and MLP to score the nodes from structural and feature perspectives, respectively and then performs fusion operations on the multi-view information scores of the nodes. In addition, to select the optimal pooling ratio of top-k, we propose a Deep-Q-Network-based Top-k (DTop-k) node selection method, which can adaptively select the best pooling ratio without prior knowledge. Experimental results on six TUDatasets and two Benchmarking GNNs datasets demonstrate the effectiveness of our proposed approach.

Transcription factors (TFs) play a part in gene expression. TFs can form complex gene expression regulation system by combining with DNA. Thereby, identifying the binding regions has become an indispensable step for understanding the regulatory mechanism of gene expression. Due to the great achievements of applying deep learning (DL) to computer vision and language processing in recent years, many scholars are inspired to use these methods to predict TF binding sites (TFBSs), achieving extraordinary results. However, these methods mainly focus on whether DNA sequences include TFBSs. In this paper, we propose a fully convolutional network (FCN) coupled with refinement residual block (RRB) and global average pooling layer (GAPL), namely FCNARRB. Our model could classify binding sequences at nucleotide level by outputting dense label for input data. Experimental results on human ChIP-seq datasets show that the RRB and GAPL structures are very useful for improving model performance. Adding GAPL improves the performance by 9.32% and 7.61% in terms of IoU (Intersection of Union) and PRAUC (Area Under Curve of Precision and Recall), and adding RRB improves the performance by 7.40% and 4.64%, respectively. In addition, we find that conservation information can help locate TFBSs.

In a traditional antenna array direction finding system, all the antenna sensors need to work or shut down at the same time, which often leads to signal crosstalk, signal distortion, and other electromagnetic compatibility problems. In addition, the direction-finding algorithm in a traditional system needs a tremendous spectral search, which consumes considerable time. To compensate for these deficiencies, a reconfigurable antenna array direction finding system is established in this paper. This system can dynamically load part or all of the antennas through microwave switches (such as a PIN diode) and conduct a fast direction of arrival (DOA) search. First, the hardware structure of the reconfigurable antenna is constructed. Then, based on the conventional spatial domain search algorithm, an improved transform domain (TD) search algorithm is proposed. The effectiveness of the system has been proven by real experiments, and the advantage of the system has been verified by detailed simulations.