Science China Information Sciences最新文献

With the growing demand for automation in agriculture, industries increasingly rely on drones to perform crop monitoring and surveillance. In this regard, fixed-wing unmanned aerial systems (UASs) are viable platforms for scanning a large crop field, given their payload capacity and range. To achieve maximum coverage without landing for battery replacement, an algorithm for producing a minimal required energy survey path is essential. Hence, an energy-aware coverage path planning algorithm is proposed herein. The constraints for a fixed-wing UAS to fly at low altitudes while achieving full coverage of the crop field are first analyzed. Then, the full path is decomposed into straight-line and U-turn primitives. Finally, an algorithm to calculate a combination of straight-line segments and U-turns is proposed to obtain the path with minimum required energy consumption. The genetic algorithm is used to efficiently determine the order of the straight-line paths to traverse. Case studies show that the proposed algorithm can produce planning results for a convex-polygon-shaped crop field.

A multispectral imaging system often cannot capture 3D spatial information owing to hardware limitations, which diminishes the effectiveness across various domains. To address this problem, we have developed a multispectral stereo imaging system along with an adaptive 3D reconstruction algorithm. Unlike existing unmanned aerial vehicle stereo imaging systems, our multispectral stereo imaging system uses two multispectral cameras with asymmetric spectral bands positioned at different angles. This design enables the acquisition of a higher number of bands and lateral spatial information while maintaining a lightweight structure. This system introduces challenges such as large geometric distortions and intensity differences between multiple bands. To accurately recover 3D spatial information, we propose an adaptive 3D reconstruction method. This method employs a position and orientation system-assisted projection transformation and a normalized threshold adjustment strategy. Finally, mutual information is used to reconstruct the multispectral images densely, effectively addressing nonlinear differences and generating a comprehensive multispectral point cloud. Our stereo system was used for two real data collections in different regions, and the efficacy of the proposed 3D reconstruction method was validated by comparing it with existing methods and commercial software.

Racing drones have attracted increasing attention due to their remarkable high speed and excellent maneuverability. However, autonomous multi-drone racing is quite difficult since it requires quick and agile flight in intricate surroundings and rich drone interaction. To address these issues, we propose a novel autonomous multi-drone racing method based on deep reinforcement learning. A new set of reward functions is proposed to make racing drones learn the racing skills of human experts. Unlike previous methods that required global information about tracks and track boundary constraints, the proposed method requires only limited localized track information within the range of its own onboard sensors. Further, the dynamic response characteristics of racing drones are incorporated into the training environment, so that the proposed method is more in line with the requirements of real drone racing scenarios. In addition, our method has a low computational cost and can meet the requirements of real-time racing. Finally, the effectiveness and superiority of the proposed method are verified by extensive comparison with the state-of-the-art methods in a series of simulations and real-world experiments.

Emotion recognition in conversation (ERC) has attracted growing attention in recent years as a result of the advancement and implementation of human-computer interface technologies. In this paper, we propose an emotional inertia and contagion-driven dependency modeling approach (EmotionIC) for ERC tasks. Our EmotionIC consists of three main components, i.e., identity masked multi-head attention (IM-MHA), dialogue-based gated recurrent unit (DiaGRU), and skip-chain conditional random field (SkipCRF). Compared to previous ERC models, EmotionIC can model a conversation more thoroughly at both the feature-extraction and classification levels. The proposed model attempts to integrate the advantages of attention- and recurrence-based methods at the feature-extraction level. Specifically, IMMHA is applied to capture identity-based global contextual dependencies, while DiaGRU is utilized to extract speaker- and temporal-aware local contextual information. At the classification level, SkipCRF can explicitly mine complex emotional flows from higher-order neighboring utterances in the conversation. Experimental results show that our method can significantly outperform the state-of-the-art models on four benchmark datasets. The ablation studies confirm that our modules can effectively model emotional inertia and contagion.

High-energy radiation detection and imaging technology has significant applications in high-energy physics research, medical imaging, and industrial monitoring. Lead-free metal halides exhibit exceptional potential for conducting indirect detection of high-energy radiation due to their characteristics of low toxicity, strong stability, high light yield, and large Stokes shift. This paper reviews the most recent advances in lead-free metal halide scintillator materials for X-ray imaging. Subsequently, it lists the most important parameters of scintillator performance and introduces the production procedures for single crystal, powder, and nanocrystal scintillators. Furthermore, it discusses the manufacturing of scintillator films with improved performance, focusing on large-area flexible scintillator films and the coupling with microstructures. Finally, it discusses current challenges and opportunities for enhancing X-ray imaging using lead-free metal halide scintillator materials.

Counterfactual explanations provide explanations by exploring the changes in effect caused by changes in cause. They have attracted significant attention in recommender system research to explore the impact of changes in certain properties on the recommendation mechanism. Among several counterfactual recommendation methods, item-based counterfactual explanation methods have attracted considerable attention because of their flexibility. The core idea of item-based counterfactual explanation methods is to find a minimal subset of interacted items (i.e., short length) such that the recommended item would topple out of the top-K recommendation list once these items have been removed from user interactions (i.e., good quality). Usually, explanations are generated by ranking the precomputed importance of items, which fails to characterize the true importance of interacted items due to separation from the explanation generation. Additionally, the final explanations are generated according to a certain search strategy given the precomputed importance. This indicates that the quality and length of counterfactual explanations are deterministic; therefore, they cannot be balanced once the search strategy is fixed. To overcome these obstacles, this study proposes learning-based counterfactual explanations for recommendation (LCER) to provide counterfactual explanations based on personalized recommendations by jointly modeling the factual and counterfactual preference. To achieve consistency between the computation of importance and generation of counterfactual explanations, the proposed LCER endows an optimizable importance for each interacted item, which is supervised by the goal of counterfactual explanations to guarantee its credibility. Because of the model’s flexibility, the trade-off between quality and length can be customized by setting different proportions. The experimental results on four real-world datasets demonstrate the effectiveness of the proposed LCER over several state-of-the-art baselines, both quantitatively and qualitatively.

Securing a comfortable, wearable compact frequency beam scanning antenna (FBSA) with robustness to deformation, low specific absorption rate (SAR), and good coverage of the surrounding environment for Internet of Things (IoT) applications, such as on-body navigation and wireless communication is an emerging challenge. In this work, a robust textile-based spoof plasmonic frequency scanning antenna utilizing higher-order modes is presented, which is also robust to deformation caused by the activities of the human body. The innovative design of the element ensures the high-efficiency transmission of the fundamental mode of spoof surface plasmon polaritons (SSPP) structure, providing the potential of being a multifunctional composite device in the compact on-body network. Besides, an artificial magnetic conductor (AMC) is designed underneath the SSPP structure, obtaining a low SAR value (0.113 W/kg), which ensures the safety of users. As a practical realization of this concept, a textile-based spoof plasmonic antenna was fabricated in the microwave regime and the performed experimental results show the proposed antenna has a single-beam radiation characteristic with a 70° beam scanning angle range when the frequency is 4.7–6.0 GHz with a high average realized gain of 13.15 dBi. And it still maintains a steady performance when faced with structure deformation, which proves its robustness. Wireless communication quality experiments are performed to demonstrate the proposed antenna can measure the angles of targets and realize wireless signal transmission to specific targets as the frequency varies, it may find great potential in the field of on-body IoT applications.

During short-range air combat involving unmanned aircraft vehicle (UAV) swarms, UAVs must make accurate maneuver decisions based on information from both enemy and friendly UAVs. This dual requirement of competition and cooperation presents a significant challenge in the field of unmanned air combat. In this paper, a method based on multi-agent reinforcement learning (MARL) is proposed to address this issue. An actor network containing three subnetworks that can handle different types of situational information is designed. Hence, the results from simpler one-on-one scenarios are leveraged to enhance the complex swarm air combat training process. Separate state spaces for local and global information are designed for the actor and critic networks. A detailed reward function is proposed to encourage participation. To prevent lazy participants in air combat, a reward assignment operation is applied to distribute these dense rewards. Simulation testing and ablation experiments demonstrate that both the transfer operation and reward assignment operation can effectively deal with the swarm air combat scenario, and reflect the effectiveness of the proposed method.

The objective of the invariant ellipsoid method is to minimize the smallest invariant and attractive set of a linear control system operating under the influence of bounded external disturbances. This study extends the application of this method to address the leader-following consensus problem. Initially, a linear control protocol is designed for the multi-agent system in the absence of disturbances. Subsequently, in the presence of bounded disturbances, by employing a similar linear control protocol, a necessary and sufficient condition is introduced to derive the optimal control parameters for the multi-agent system such that the state of followers converges to and remains in a minimal invariant ellipsoid around the state of the leader.

We present a wavelet-domain feature-decoupling Transformer-based tracking network for the weakly supervised MOT task (FDMOT). Our FDMOT has two improvements over the previous weakly supervised methods. First, FDMOT decouples noisy intermediate features caused by noisy pseudo identity labels in the wavelet domain, extracting discriminative features for accurately detecting and identifying multiple targets. Second, FDMOT further improves the noise-decoupled embedding features into the well-refined ones with the cooperation of the three feature-decoupling Transformer-based branches, which can accurately identify and track heavily occluded targets in crowded scenes. Experimental results show the superiority of FDMOT compared with several state-of-the-art supervised and weakly supervised MOT methods.