Computer-aided diagnosis (CAD) has nearly fifty years of history and has assisted many clinicians in the diagnosis. With the development of technology, recently, researches use the deep learning method to get high accuracy results in the CAD system. With CAD, the computer output can be used as a second choice for radiologists and contribute to doctors doing the final right decisions. Chest abnormality detection is a classic detection and classification problem; researchers need to classify common thoracic lung diseases and localize critical findings. For the detection problem, there are two deep learning methods: one-stage method and two-stage method. In our paper, we introduce and analyze some representative model, such as RCNN, SSD, and YOLO series. In order to better solve the problem of chest abnormality detection, we proposed a new model based on YOLOv5 and ResNet50. YOLOv5 is the latest YOLO series, which is more flexible than the one-stage detection algorithms before. The function of YOLOv5 in our paper is to localize the abnormality region. On the other hand, we use ResNet, avoiding gradient explosion problems in deep learning for classification. And we filter the result we got from YOLOv5 and ResNet. If ResNet recognizes that the image is not abnormal, the YOLOv5 detection result is discarded. The dataset is collected via VinBigDatas web-based platform, VinLab. We train our model on the dataset using Pytorch frame and use the mAP, precision, and F1-score as the metrics to evaluate our models performance. In the progress of experiments, our method achieves superior performance over the other classical approaches on the same dataset. The experiments show that YOLOv5s mAP is 0.010, 0.020, 0.023 higher than those of YOLOv5, Fast RCNN, and EfficientDet. In addition, in the dimension of precision, our model also performs better than other models. The precision of our model is 0.512, which is 0.018, 0.027, 0.033 higher than YOLOv5, Fast RCNN, and EfficientDet.

In recent years, sparse coding has been used in a wide range of applications including classification and recognition. Different from many other applications, the sparsity pattern of features in many classification tasks are structured and constrained in some feasible domain. In this paper, we proposed a re-weighted \(\ell _{2,1}\) norm based structured sparse coding method to exploit such structures in the context of classification and recognition. In the proposed method, the dictionary is learned by imposing the class-specific structured sparsity on the sparse codes associated with each category, which can bring noticeable improvement on the discriminability of sparse codes. An alternating iterative algorithm is presented for the proposed sparse coding scheme. We evaluated our method by applying it to several image classification tasks. The experiments showed the improvement of the proposed structured sparse coding method over several existing discriminative sparse coding methods on tested data sets.

The Bearing–Angle algorithm effectively improves the observability of vision-based motion estimation for moving targets by combining the dimensional information of target detection frames. However, the robustness of this algorithm will be significantly reduced when the observation error increases due to sudden changes in the target motion state. To address this shortcoming, this paper proposes a visual target motion estimation algorithm called the Dynamic Bearing–Angle, which aims to improve the accuracy and robustness of target motion analysis in dynamic scenarios such as unmanned aerial vehicle (UAV). The algorithm innovatively introduces a dual robustness mechanism of dynamic noise intensity adaptation and outlier suppression based on M-estimation. By adjusting the noise covariance matrix in real time and assigning low weights to the outlier observations using the Huber weight function, the Dynamic Bearing–Angle algorithm is able to effectively cope with non-Gaussian noise and sudden target maneuvers. We validate the performance of the proposed algorithm with numerical simulations and real sensor data, and the results show that the Dynamic Bearing–Angle maintains good robustness and accuracy under different noise intensities.

Limited computational resources prevent GNSS-based passive radar systems from processing all accessible signals, necessitating intelligent signal selection for efficient target tracking. This paper proposes an adaptive selection method based on Rényi divergence. Within the Cardinality Balanced Multi-Bernoulli (CBMeMBer) filter framework, the method establishes an optimization model that maximizes the expected information gain under a fixed signal-number constraint. To comprehensively validate performance, simulations are conducted under three scenarios: multi-target linear motion, single-target tracking (for comparison with the classical Geometric Dilution of Precision (GDOP) criterion), and multi-target nonlinear maneuvering. Results demonstrate that the proposed algorithm significantly reduces computational load while achieving tracking accuracy superior to random selection and comparable to using all satellites. Compared to the GDOP-based method, it exhibits improved steady-state tracking accuracy by leveraging its dynamic, information-driven selection mechanism. This work provides an effective solution for intelligent resource management in resource-constrained GNSS-based passive radar systems.

Estimating the relative position and velocity of micro aerial vehicles (MAVs) using visual signals is a critical issue in numerous tasks. However, traditional relative motion estimation algorithms suffer severely from non-Gaussian noise interference and have limited observability, making it difficult to meet the practical requirements of complex dynamic scenarios. To address this dilemma, this paper proposes a Multimodal Decoupled Spatiotemporal Adaptive Network (MDSAN). Designed for air-to-air scenarios, MDSAN achieves high-precision relative pose and velocity estimation of dynamic MAVs while overcoming the observability limitations of traditional algorithms. In detail, MDSAN is collaboratively composed of two core sub-modules: Modality-Specific Convolutional Normalization (MSCN) blocks and Spatiotemporal Adaptive State (STAS) blocks. Specifically, MSCN uses custom convolution kernels tailored to three modalities—visual, physical, and geometric—to separate their features. This prevents interference between modalities and reduces non-Gaussian noise. STAS, built on a state-space model, combines two key functions: it tracks long-term MAV motion trends over time and strengthens the synergy between different modal features across space. Adaptive weights balance these two functions, enabling stable estimation, even when traditional methods struggle with low observability. Furthermore, MDSAN adopts a full-vision multimodal fusion scheme, completely eliminating the dependence on wireless communication and reducing hardware costs. Extensive experimental results demonstrate that MDSAN achieves the best performance in all scenarios, significantly outperforming existing motion estimation algorithms. It provides a new technical path that balances high precision, high robustness, and cost-effectiveness for technologies such as MAV swarm perception.

Vision‐based unmanned aerial vehicle (UAV) tracking technology faces severe challenges in dynamic and complex air‐to‐air scenarios, manifested specifically as occlusion interference, abrupt changes in light intensity, and feature motion blur caused by high‐speed target maneuvering. Traditional tracking methods are limited by static template strategies and fail to effectively model the temporal correlation of motion cues, which easily leads to trajectory drift and even target loss. To address the above bottlenecks, this paper proposes a real‐time UAV target tracking algorithm optimized based on the dual memory template‐motion correction (DMT‐MC) strategy. The algorithm adopts a dual memory template strategy to balance the long‐term stability and short‐term dynamic adaptability of target features. Meanwhile, it introduces adaptive Kalman filtering guided by matching scores for motion correction, fusing spatiotemporal cues to improve the tracking stability of the target in scenarios such as fast movement and occlusion, and avoiding trajectory jumps caused by single‐frame feature noise. To verify the algorithm performance, this paper constructs A2A‐UAV‐Tracking, the first benchmark dataset for air‐to‐air UAV tracking in highly adversarial environments. Comprehensive experimental results demonstrate that the target tracking algorithm optimized by the DMT‐MC strategy significantly outperforms other mainstream lightweight trackers used for comparison, fully proving its effectiveness and robustness in complex dynamic scenarios. The constructed dataset has been made available for download at https://github.com/KingArthurYu/A2A-UAV-Tracking .