Explore the vast range of titles available.

In recent years, demand has been increasing for target detection and tracking from aerial imagery via drones using onboard powered sensors and devices. We propose a very effective method for this application based on a deep learning framework. A state-of-the-art embedded hardware system empowers small flying robots to carry out the real-time onboard computation necessary for object tracking. Two types of embedded modules were developed: one was designed using a Jetson TX or AGX Xavier, and the other was based on an Intel Neural Compute Stick. These are suitable for real-time onboard computing power on small flying drones with limited space. A comparative analysis of current state-of-the-art deep learning-based multi-object detection algorithms was carried out utilizing the designated GPU-based embedded computing modules to obtain detailed metric data about frame rates, as well as the computation power. We also introduce an effective target tracking approach for moving objects. The algorithm for tracking moving objects is based on the extension of simple online and real-time tracking. It was developed by integrating a deep learning-based association metric approach with simple online and real-time tracking (Deep SORT), which uses a hypothesis tracking methodology with Kalman filtering and a deep learning-based association metric. In addition, a guidance system that tracks the target position using a GPU-based algorithm is introduced. Finally, we demonstrate the effectiveness of the proposed algorithms by real-time experiments with a small multi-rotor drone.

In response to the challenges of target recognition and misjudgment caused by varying target scales, diverse shapes, and interference such as lake surface reflections in river and lake scenarios, this paper proposes the YOLO v11n-DDH model for fast and detection of spatial targets in river and lake environments. The model builds upon YOLO v11n by introducing the Dynamic Snake Convolution (DySnakeConv) to enhance the ability to extract detailed features. It integrates the Deformable Attention Mechanism (DAttention) to strengthen key features and suppress noise, while combining the improved High-Level Screening Feature Pyramid Network (HSFPN) structure for multi-level feature fusion, thus improving the semantic representation of targets at different scales. Experiments on a self-constructed dataset show that the precision, recall, and mAP of the YOLO v11n-DDH model reached 88.4%, 78.9%, and 83.9%, respectively, with improvements of 3.4, 2.9, and 2.5 percentage points over the original model. Specifically, DySnakeConv increased mAP@50 by 0.6 percentage points, DAttention improved mAP@50 by 0.3 percentage points, and HSFPN contributed to a 0.9 percentage point rise in mAP@50. This patrol system can effectively identify and visualize various pollutants in river and lake areas, such as underwater waste, water quality pollution, illegal swimming and fishing, and the “Four Chaos” issues, providing technical support for intelligent river and lake management.

UAV technology is a basic technology aiming to help realize smart living and the construction of smart cities. Its vigorous development in recent years has also increased the presence of unmanned aerial vehicles (UAVs) in people’s lives, and it has been increasingly used in logistics, transportation, photography and other fields. However, the rise in the number of drones has also put pressure on city regulation. Using traditional methods to monitor small objects flying slowly at low altitudes would be costly and ineffective. This study proposed a real-time UAV tracking scheme that uses the 5G network to transmit UAV monitoring images to the cloud and adopted a machine learning algorithm to detect and track multiple targets. Aiming at the difficulties in UAV detection and tracking, we optimized the network structure of the target detector yolo4 (You Only Look Once V4) and improved the target tracker DeepSORT, adopting the detection-tracking mode. In order to verify the reliability of the algorithm, we built a data set containing 3200 pictures of four UAVs in different environments, conducted training and testing on the model, and achieved 94.35% tracking accuracy and 69FPS detection speed under the GPU environment. The model was then deployed on ZCU104 to prove the feasibility of the scheme.

Ensuring the welfare of dairy cows requires precise monitoring of their daily exercise to evaluate their physical health. This necessitates innovative methods beyond traditional motion sensors. We present a novel method that integrates an enhanced YOLOv5s object detection model with the DeepSORT multi-object tracking algorithm to meticulously track dairy cow movements, providing holistic information about their health. Our research started with the establishment of a dedicated dataset tailored for cow detection. We then segmented the detection scope to focus on specific regions of interest. Within the modified YOLOv5s model, we replaced the standard CSPDarknet53 backbone with DenseNet to achieve deep separable convolution and feature reorganization modules, leading to reduced parameters, augmented feature expression, and better information flow. In particular, the SPD-Conv module was incorporated to retain intricate details, essential for detecting smaller and low-resolution targets. The transition from Generalized Intersection over Union (GIoU) Loss to Complete Intersection over Union (CIoU) loss improved detection accuracy and sped up model convergence. Our clustering approach, based on the elbow rule, optimized K-means clustering, enhancing speed and precision. For multi-object tracking, the DeepSORT model was tailored to cater to varying cow sizes, and we chose an algorithm to associate appearance information. We converted pixel data into real-world distances, providing exact 3D cow movement metrics. Experimental validation confirmed the efficacy of our approach. Our enhanced model surpassed the original YOLOv5s in performance by 11.1% for accuracy (97.4%), 9.6% for recall (97.8%), and 11.0% for average accuracy (98.2%). The comprehensive accuracy stood at 92.1% for our model. In conclusion, our innovative methodology offers a non-invasive means to monitor dairy cow exercise, paving the way for advanced health assessment techniques in the dairy sector.

Sports video analysis has produced many valuable applications driven by different needs, and in these applications, moving target detection technology plays an indispensable role. However, the uniqueness of sports videos brings a big challenge to target detection and tracking technology. Therefore, the purpose of this article is to propose an efficient multi-target detection algorithm to quickly and effectively detect all target objects in the video. We propose a multi-target detection and tracking framework based on a deep conditional random field network, adding a conditional random field layer to the output of the target detection network to model the mutual relationships and contextual information between targets. In addition, we also introduce local adaptive filters and spatial-temporal attention mechanisms into this framework to further improve target detection performance, especially when dealing with complex scenes and target interactions. Experimental results show that the proposed method is superior to the state-of-the-art methods in terms of accuracy and efficiency.

Sidelobe suppression is a common and critical issue in radar pulse compression processing. However, the traditional window functions such as Hamming and Hanning windows are all limited to fixed weight parameters, and can only be applied to a specific application. Thus, we propose a flexible window function design algorithm based on chaotic particle swarm optimization (W-CPSO). First, a polynomial representation of the cosine-sum window functions is derived using Taylor series expansion. Then, the CPSO-based optimization algorithm is proposed to improve the sidelobe suppression performance of window functions by optimizing the polynomial coefficients. The proposed algorithm has been applied to optimize several widely used windows, including the Hamming, Hann, Blackman windows, etc. Analysis results demonstrate that the optimized window functions significantly improve sidelobe suppression while maintaining comparable mainlobe width. Furthermore, the simulation results of multi-target detection in radar systems also validate the effectiveness of the proposed algorithm, showing that the optimized windows exhibit superior performance in detecting weak targets compared to the traditional windows.

This research focuses on the problem of multi-target detection and multi-bomb compound target fire distribution in the key field of military operations, and explores the solution based on numerical calculation method. For multi-target detection, a high-precision detection model combining deep learning and mathematical morphology optimization is constructed. An innovative strategy based on mixed integer programming and quantum genetic algorithm is proposed for fire allocation of multi-projectile compound targets. Through rigorous mathematical derivation and model construction, the accuracy and speed of multi-target detection are significantly improved, and the efficient allocation of resources and the maximization of combat effectiveness are achieved in fire allocation. The experimental results show that the accuracy of the multi-target detection model is significantly improved compared with the traditional method, and the multi-weapon composite target fire allocation scheme has obvious advantages in resource utilization and target damage effect, which can provide accurate technical support for military combat decision-making.

For a typical FMCW automotive radar system, a new design of baseband signal processing architecture and algorithms is proposed to overcome the ghost targets and overlapping problems in the multi-target detection scenario. To satisfy the short measurement time constraint without increasing the RF front-end loading, a three-segment waveform with different slopes is utilized. By introducing a new pairing mechanism and a spatial filter design algorithm, the proposed detection architecture not only provides high accuracy and reliability, but also requires low pairing time and computational loading. This proposed baseband signal processing architecture and algorithms balance the performance and complexity, and are suitable to be implemented in a real automotive radar system. Field measurement results demonstrate that the proposed automotive radar signal processing system can perform well in a realistic application scenario.

This paper presents an enhanced YOLOv8 model designed to address multi-target detection challenges in complex traffic scenarios. The model integrates the Squeeze-and-Excitation attention mechanism, the deformable convolution C2f module, and the smooth IoU loss function, achieving significant improvements in detection accuracy and robustness in various complex environments. Experimental results show that the enhanced YOLOv8 model outperforms existing YOLO solutions across multiple metrics, particularly in precision and recall. Specifically, the enhanced model achieves 83.8% precision and 82.7% recall, improving 1.05 times in precision and 1.1 times in recall compared to the average precision (79.7%) and recall (75.4%) of other YOLO series models. In terms of mAP₀.5, the enhanced model achieves 89%, representing a 1.05-fold improvement over the average mAP₀.5 (84.4%) of YOLO series models. For mAP₀.5:0.95, the enhanced model reaches 76.5%, which is a 1.1-fold improvement over the average mAP₀.5:0.95 (69.7%) of YOLO series models. These improvements demonstrate the superior performance of the proposed model in multi-scale and complex scenarios, providing strong support for intelligent transportation systems and autonomous driving.

To address the challenges of radar multi-target detection in marine environments, this paper proposes an information geometry (IG)-based, two-stage track-before-detect (TBD) framework. Specifically, multi-target measurements are first modeled on the manifold, leveraging its geometric properties for enhanced detection. The designed scoring function incorporates both the feature dissimilarity between targets and clutter, as well as the precise inter-target path associations. Consequently, a novel merit function combining feature dissimilarity and transition cost is derived to mitigate the mutual interference between adjacent targets. Subsequently, to overcome the integrated merit function expansion phenomenon, a two-stage integration strategy combining dynamic programming (DP) and greedy integration (GI) algorithms was adopted. To tackle the challenges of unknown target numbers and computationally infeasible multi-hypothesis testing, a target cancellation detection scheme is proposed. Furthermore, by exploiting the independence of multi-target motions, an efficient implementation method for the detector is developed. Experimental results demonstrate that the proposed algorithm inherits the superior clutter discrimination capability of IG detectors in sea clutter environments while effectively resolving track mismatches between neighboring targets. Finally, the effectiveness of the proposed method was validated using real-recorded sea clutter data, showing significant improvements over conventional approaches, and the signal-to-clutter ratio was improved by at least 2 dB.