Detecting pavement distress in a timely manner is crucial for urban safety and higher traffic efficiency. Owing to their high accuracy and real-time capabilities, object detection methods have emerged as effective solutions for road condition monitoring. However, striking a balance between detection performance and model lightweighting remains a significant challenge. To address this, we propose a novel pavement distress detection model based on YOLOv11n. First, the Ghost module is employed to significantly reduce parameters while preserving detection performance. Then, a novel and versatile high-resolution feature extraction module (HR) is designed to enhance spatial feature extraction and improve the utilization of high-resolution features. In addition, two dedicated feature extraction modules are introduced to optimize gradient flow and expand the receptive field. They enable more accurate detection of detailed damage. To further enhance multi-scale feature integration, an improved fusion structure is incorporated, where max pooling is used to effectively aggregate contextual information. Experimental results demonstrate that the proposed model improves detection accuracy, with mAP50 increasing from 71.3% to 72.9%, and mAP50-95 rising from 42.3% to 43.9%. Meanwhile, the model maintains a lightweight design, reducing parameters and model size by 13.57% and 8.81%, respectively-resulting in only 2.23M parameters and a 4.76MB model size. In addition, the model achieves an average inference time of 8.8 ms per image, enabling real-time performance. Moreover, it shows strong generalization ability on the real-world Nanjing urban dataset, achieving a competitive trade-off between accuracy and efficiency, making it suitable for deployment in resource-constrained environments.

Drones are widely used in traffic management due to their wide inspection range and high flexibility. Among them, target detection based on drone images plays an important role. However, target detection using drones faces the problem of small targets and insufficient detailed information in the image, resulting in low accuracy of model detection and a high rate of false detection and missed detection. This paper proposes an improved small object detection algorithm, S-YOLOv11, to solve the above problems. First, an enhanced multi-branch auxiliary feature pyramid network (EMAFPN) is designed in the neck, inspired by the ideas of multi-branch auxiliary feature pyramid network (MAFPN) and bidirectional feature pyramid network (BiFPN) to strengthen the ability of effective feature information exchange and fusion between networks. Second, by introducing an efficient shared detail-enhanced convolutional detection head (ESDCDH), the model’s ability to locate, classify, and capture details is effectively improved. Finally, the learning ability of the model to small size and difficult-to-detect objects is enhanced by using the Intersection over Union with Normalized Wasserstein Distance and Minimum Point Distance (NWD-MPD-IoU) loss function. Experimental results demonstrate that compared with the original model, the S-YOLOv11 model improves mAP0.5, mAP0.5:0.95, P, and R on the public dataset VisDrone2019 by 11.8

In recent years, Siamese-based trackers have achieved promising results in visual object tracking. However, the feature extraction capability of current popular Siamese-like networks is limited, making it difficult to fully distinguish the object from the background. Trackers are susceptible to drifting caused by factors such as occlusion, scale variation, and fast motion. In this paper, we propose a novel tracker, dubbed Siamese network with asymmetric target-aware and task-aware (SiamATA). The network is based on the asymmetric structure of the classification-regression branches, including the template classification branch, template regression branch, search region classification branch, and search region regression branch, to alleviate overfitting. Meanwhile, a target-aware attention module is introduced to learn powerful context information through spatial attention and selectively emphasize dependency channel features through channel attention, providing target-aware semantic features for each branch. In addition, we adopt the nonlocal pixel-wise correlation method to suppress the influence of similar object interference. Finally, we design a frequency domain task-aware attention module to explore the self-semantic information of classification and regression branches. Extensive experiments demonstrate the effectiveness of our tracker on six benchmarks: OTB100, UAV123, VOT2018, VOT2019, GOT-10K, and LaSOT.

The implementation of Scenario Classification (SC) plays a pivotal role in various edge intelligence applications, notably in fields such as autonomous driving, navigation, and remote sensing. With recent advancements, deep learning (DL) techniques have substantially improved SC, delivering remarkable results in classification tasks. However, the integration of DL in SC brings significant computational demands, posing challenges for deployment on edge devices where resources are constrained. Addressing this issue, we propose a novel Low-Resource Scenario Classification (LR-SC) approach, primarily focused on model pruning. This strategy aims to reduce computational power and storage needs, thus optimizing resource utilization in edge intelligence applications. Our approach involves the application of an ℓ2 regularization and a threshold-based pruning method, which selectively eliminates non-essential connections. This is followed by a systematic process of alternating pruning and fine-tuning to mitigate any performance loss due to the pruning. Experimental evaluations of the LR-SC method have shown its effectiveness; it substantially lowers the parameter count to merely 24% of the original model, while simultaneously achieving a 0.42% increase in classification accuracy.

Unsupervised domain adaptation (UDA) aims to extract domain-invariant features. Existing UDA methods mainly utilize a convolutional neural network (CNN) or vision transformer (ViT) as feature extractor, which align in the latent space that characterizes single view of the object and may lead to matching error—the distributions aligned in the CNN space may still be confused in the ViT space. To address this, we introduce global–local bi-alignment (GLBA) based on a hybrid structure Conformer, which enforces simultaneous alignment in both spaces, following the space-independent assumption: If two domains have the same distribution, their distributions in any latent space are aligned. The framework can be easily combined with previous UDA methods, essentially adding only an alignment loss without the need for elaborate structures or large numbers of parameters. Experiments demonstrate the effectiveness of GLBA and its state-of-the-art (SoTA) performance achieved with comparable parameter complexity. Code is available at https://github.com/JSJ515-Group/GLBA .

Based on the systematic deconstruction of multi-dimensional and multi-target biological networks,modular pharmacology explains the complex mechanism of diseases and the interactions of multi-target drugs.It has made progress in the fields of pathogenesis of disease,biological basis of disease and traditional Chinese medicine(TCM)syndrome,pharmacological mechanism of multi-target herbs,compatibility of formulas,and discovery of new drug of TCM compound.However,the complexity of multi-omics data and bio-logical networks brings challenges to the modular deconstruction and analysis of the drug networks.Here,we constructed the"Compu-ting Platform for Modular Pharmacology"online analysis system,which can implement the function of network construction,module identification,module discriminant analysis,hub-module analysis,intra-module and inter-module relationship analysis,and topologi-cal visualization of network based on quantitative expression profiles and protein-protein interaction(PPI)data.This tool provides a powerful tool for the research on complex diseases and multi-target drug mechanisms by means of modular pharmacology.The platform may have broad range of application in disease modular identification and correlation mechanism,interpretation of scientific principles of TCM,analysis of complex mechanisms of TCM and formulas,and discovery of multi-target drugs.

Object detection includes three subtasks of predicting target position, classification, and confidence. In the mainstream object detection model, the model pursues internal structure refinement, and each subtask shares almost the same structure, which is a task-coupled structure. The task-coupled structure of the model reduces the training parameters, but it cannot be tuned on the network structure for each task separately, which can limit the model performance. We designed a task decoupled object detection network (YOLOD) based on YOLOv5, where YOLOD is decoupled immediately after the backbone network. By observing the loss convergence of each subtask, three network structures are designed separately and the branch size is controlled so that the model has fewer training parameters. At the same time, some experimental adjustments were made to YOLOD to accelerate the convergence speeds of the model. In addition, we add image contour information to the original three-channel image to assist model training and improve detection accuracy. The experiments demonstrate that the modified model is smaller in size and has the largest accuracy improvement on the small-scale model. without introducing any attention-based modules, YOLOD-S achieves a mAP improvement of 1.1% on the MS COCO dataset and 2.29% on the VOC dataset, and the larger model YOLOD-L achieves an accuracy of 48.8% on the COCO dataset.

Online knowledge distillation breaks the pre-determined strong and weak teacher-student models, it provides a new way of thinking about knowledge distillation. However, the current online methods often use the Logits-based prediction distribution, and the features containing rich semantic information are rarely used. Even if the feature-based methods are used, they only operate on the last layer of the network, without further exploring the representation knowledge of the middle layer feature map. To address the above issues, we propose an innovative feature early fusion and reconstruction (FEFR) method for online knowledge distillation which entails four essential components: multi-scale feature extraction and intermediate layer feature early fusion, reconstruction of features, dual-attention and overall fusion module in this paper. We propose early fusion by “sum” operation for feature matrices between different layers and advance fusion to improve the feature map representation. In order to enhance the communication ability between groups to obtain features, the features were reconstructed. We create a dual-attention to enhance the critical channel and spatial regions adaptively in order to collect more accurate information. The previously processed feature maps are combined and fused using feature fusion, which also aids in student models training. A study of the network architectures of CIFAR-10, CIFAR-100, CINIC-10 and ImageNet 2012 shows that FEFR provides more useful characterization knowledge for refinement and improves accuracy by about 0.5% compared to other methods.

The premise of effective use of high-resolution remote sensing (HRRS) images is that the data integrity and authenticity of HRRS images must be guaranteed. This paper proposes a new subject-sensitive hashing algorithm for the integrity authentication of HRRS images. This algorithm takes AGIM-net (Attention Gate-based improved M-net) proposed in this paper to extract the subject-sensitive features of the HRRS images, and uses Principal Component Analysis (PCA) based method to compress and encode the extracted features. AGIM-net is an improved U-net based on attention mechanism, adding multi-scale input in the encoder stage to extract rich image features; adding multi-scale output in the decoder stage, and suppressing the features irrelevant to the subject through Attention Gate to improve the robustness of the algorithm. Experiments show that the proposed algorithm has improved robustness compared with existing algorithms, and the tamper sensitivity and security are basically equivalent to the existing algorithms.

Knowledge distillation is an emerging method for acquiring efficient, small-scale networks. The main idea is to transfer knowledge from a complex teacher model with high learning capacity to a simple student model. To this end, various approaches to knowledge distillation have been proposed in the past few years, focusing mainly on modifications to student learning styles and less on changes to teacher teaching styles. Therefore, our new approach to knowledge distillation teacher training involves adapting the trained teachers to the knowledge distillation model in order to minimize the gap between the student model and the teacher model. We introduced the idea of a "Trained Teacher": Our approach involves using a specially trained teacher network that, by incorporating knowledge distillation constraints during its own training, adapts to the teaching model in advance and performs nearly identically to a typical teacher network. This allows students to absorb the teacher's knowledge more effectively, thereby increasing their competence. In addition, the methods of mainstream knowledge distillation currently in use are equally appropriate to our educated teachers. Extensive tests on numerous datasets reveal that our technique outperforms the original knowledge distillation in accuracy on standard KD by 2%. Our code and pre-trained models can be found at https://github.com/JSJ515-Group/Traine d_teacher.

The implicit prerequisite for using HRRS images is that the images can be trusted. Otherwise, their value would be greatly reduced. As a new data security technology, subject-sensitive hashing overcomes the shortcomings of existing integrity authentication methods and could realize subject-sensitive authentication of HRRS images. However, shortcomings of the existing algorithm, in terms of robustness, limit its application. For example, the lack of robustness against JPEG compression makes existing algorithms more passive in some applications. To enhance the robustness, we proposed a Transformer-based subject-sensitive hashing algorithm. In this paper, first, we designed a Transformer-based HRRS image feature extraction network by improving Swin-Unet. Next, subject-sensitive features of HRRS images were extracted by this improved Swin-Unet. Then, the hash sequence was generated through a feature coding method that combined mapping mechanisms with principal component analysis (PCA). Our experimental results showed that the robustness of the proposed algorithm was greatly improved in comparison with existing algorithms, especially the robustness against JPEG compression.

针对高速公路小目标检测召回率低、准确率低等问题,提出一种基于YOLOv5改进的高速公路小目标检测算法.针对数据集样本不均衡问题,引入克隆数据增强对数据集进行增广;通过K-means算法以IOU作为度量值,聚类产生适合该文数据集的Anchors boxes;通过添加注意力模块,加强通道注意力,降低模型噪声并提高准确率;针对检测小目标困难问题,引入三层特征融合机制,加强模型对浅层信息特征的提取能力;通过增加检测头,提高对小目标特征的提取能力.实验结果表明,改进后的YOLOv5模型在自建数据集上mAP达到了0.68,Recall达到了0.64,对小目标有很好的检测效果.

As a research hotspot, pedestrian detection has a wide range of applications in the field of computer vision in recent years.However, current pedestrian detection methods have problems such as insufficient detection accuracy and large models that are not suitable for large-scale deployment.In view of these problems mentioned above, a lightweight pedestrian detection and early warning method using a new model called you only look once (Yolov5) is proposed in this paper, which utilizing advantages of Yolov5s model to achieve accurate and fast pedestrian recognition.In addition, this paper also optimizes the loss function of the batch normalization (BN) layer.After sparsification, pruning and fine-tuning, got a lot of optimization, the size of the model on the edge of the computing power is lower equipment can be deployed.Finally, from the experimental data presented in this paper, under the training of the road pedestrian dataset that we collected and processed independently, the Yolov5s model has certain advantages in terms of precision and other indicators compared with traditional single shot multiBox detector (SSD) model and fast region-convolutional neural network (Fast R-CNN) model.After pruning and lightweight, the size of training model is greatly reduced without a significant reduction in accuracy, and the final precision reaches 87%, while the model size is reduced to 7,723 KB.

本系统采用YOL0v5实现有害生物检测算法,手机端采用NCNN部署模型,离线数据库采用SQLite,同时实现WEB端平台,完成系统管理与可视化统计分析.通过真实环境下的测试得出,mAP值约为93％,测试精度约为71％.在安卓设备上进行识别约耗时100ms,较好地实现了功能.

文章针对"数据库系统"在某应用型高校网络工程专业中的实际教学情况,研究课程思政元素,从思政意义、研究现状、教学目标、教学方法及教学案例等方面阐述对该课程中思政元素的研究与思考.

Road safety has always been one of the main concerns. With the development of deep learning, computer vision has begun to be used in road damage detection. It has the advantages of faster detection speed, lower cost, easier deployment, etc., which greatly reduces traffic accidents caused by the road. We design an automatic detection system for road damages and deploy it on NVIDIA Jetson Xavier NX. A dataset named FocusCrack is collected under diversiform roads and various lighting conditions. It contains six types of diseases, a total of 4181 images and 5812 labels. Compared with performance of several mainstream algorithms like Faster R-CNN and Single Shot MultiBox Detector (SSD), the model adopts the You Only Look Once v5s(YOLOv5s) algorithm. After experimental testing, the precision, recall, mAP@0.5, and mAP@0.5:0.95 are 90.1%, 91.3%, 93.8%, and 51.9%. The system has achieved good results in practical application.

Keypoint-based detectors generate a large number of false positives due to incorrect keypoint matching in the object detection task. In this paper, we propose an adaptive center keypoint selection method (ACKSNet) to address the false-positive drawback. We first roughly group detected corners by associative embeddings, which flexibly localize objects of various shapes and scales to obtain a large number of initial candidate proposals. Then, ACKSNet associates introduced extra center keypoints with corner pairs through the geometric method to add location information in the candidate regions. And it independently generates a threshold for each center keypoint according to their statistical characteristics to ensure the high quality of center keypoints. Furthermore, we enrich the scale information of the output feature maps by equipping the backbone network with dilated convolution modules. On the MS COCO dataset, our model achieves an AP of 44.5%, surpassing most existing anchor-free detectors.

Vehicle position awareness is a kernel technique for the Internet-of-Vehicles (IOV). In this paper, we develop a novel vehicle localization framework, the core of which is to measure the two-dimensional (2D) angle-of-arrival (AOA) of the vehicle via coprime polarized multiple-input multiple-output (MIMO) systems. The transmitting polarized array emit mutual orthogonal training waveforms, and the receiving array pick up the waveforms via matched filters. A parallel factor (PARAFAC) decomposition algorithm is carried out, which first obtains the ambiguous elevation angles corresponding to the transmit array and receive array, and then it estimates the unambiguous elevation angles via the coprime characteristic of the arrays. Thereafter, it achieves the azimuth angles via the vector cross-product technique. Finally, the vehicle position is calculated via the relationship between the MIMO system and the vehicle. The proposed framework is suitable for anonymous vehicle. Numerical results are provided to show the effectiveness of the proposed framework.

Due to the complexity and variability of lighting conditions, weather conditions, and the location relationship between vehicles and cameras, the license plate detection in complex scenes has an important application value. In this paper, based on the full analysis of the edge features of license plate area in complex scenes, a new fusion image preprocessing method using edge features is proposed, and the rectangular box similar to license plate area is taken as the candidate box. The experimental results show that the accuracy of the location of license plate area under different conditions reaches 98.31%. Based on the result of coarse location, we propose a Bayesian classifier to realize accurate location of vehicle license plate. This classifier makes full use of the high efficiency of Bayesian method to process low dimensional data and the credibility of probability calculation distribution. The model in different conditions on achieved 91.59% precision. The results of comparison with other classifiers show that the method in this paper is very robust to license plate detection in complex scenarios.

复杂场景下的车牌定位算法由于环境的变化导致车牌识别率低、计算效率不高.在对Faster R-CNN算法分析的基础上,结合MEAN-SHIFT聚类算法的特点,提出了一种基于MEAN-SHIFT特征增强的车牌定位方法.该方法采用并行计算的方式,通过增强目标区域的特征,有效提升了复杂场景下车牌定位的效率和准确度.实验表明,该方法能够在多种复杂场景下快速定位车牌照区域,准确率高,具有较好的鲁棒性.