LoRa, a leading wireless technology for Low Power Wide Area Networks (LPWAN), is well-known for its long transmission range and low power consumption. The extended range is primarily attributed to the Chirp Spread Spectrum technique. However, the LoRa physical layer (LoRa PHY) contributes only marginally to this advantage, as it employs an inefficient Forward Error Correction (FEC) strategy for error recovery. In this paper, we introduce Sym-FEC, a symbol-level FEC decoder designed to link the received signals' spectrum with the coding correlations inherent in LoRa PHY, thereby enhancing error recovery. The key enabler of Sym-FEC is signal copy retrieval. We begin by facilitating signal copy conversion between two symbols and extend this to the general case, where signal copy conversions can be performed between any symbols in a coding block. Approaches are also introduced to assess the validity of the block-wide decoding results. Extensive hardware evaluations demonstrate that Sym-FEC provides Signal-to-Noise-Ratio (SNR) improvement of 2.3dB to 3dB compared to the traditional decoder in LoRa PHY. Sym-FEC requires no modifications at the transmitter while incurs low storage and computational complexity at the gateway, thus can be easily integrated into gateway nodes.

Seed germination rate is a critical factor in agricultural productivity. Traditional approaches to germination assessment necessitate human scrutiny, introducing subjectivity and diminishing operational efficiency. While fully supervised deep learning approaches offer objectivity, reproducibility and efficiency, they require large-scale and high-quality labeled datasets, which are often challenging to obtain. To address this limitation, this study introduce a Semi-Supervised Germination Detection (SSGD) method built upon the Soft Teacher framework. SSGD employs a Faster R-CNN detector with ResNet50-FPN feature extraction network in both the teacher model and the student model. To ensure higher learning stability, the parameters of the teacher model are dynamically improved by an exponential moving average (EMA) updating mechanism. This study conducted a comprehensive evaluation of SSGD on the publicly available Pennisetum glaucum (PG), Secale cereale (SC), and Zea mays (ZM) datasets. Remarkably, with only 10% of labeled data, SSGD achieved mAP50 scores of 0.954, 0.928, and 0.961 on PG, SC, and ZM, respectively, surpassing fully supervised methods trained on 100% labeled data, including YOLOv3, FCOS, Cascade R-CNN and Faster R-CNN. Moreover, SSGD consistently outperformed the Faster R-CNN baseline across various annotation ratios (1%, 10%, 20% and 30%). Notably, even when the PG dataset’s labeling rate dropped to only 1%, SSGD maintained a high mAP50 of 0.94, exceeding the baseline by 5 percentage points. These findings underscore SSGD’s strong adaptability to limited data and further emphasize the effectiveness of semi-supervised learning in seed germination detection.

The integration of hyperspectral imaging (HSI) and computer vision (CV) technologies enables rapid, accurate, and non-destructive seed quality detection methods, providing technical support for accurate seed classification and rational utilization of seed resources. However, in practical applications, challenges such as high computational costs and complex feature extraction remain, leading to limited modeling capability. This paper summarizes applications of CV-based HSI technology in seed quality detection, focusing on the current research status of seed quality-related models and methods, such as 1) variety classification; 2) vigor assessment; 3) moisture content determination; 4) health determination; along with the future development trends.

The flocking movement is a fundamental and crucial operation in multi-AAVs systems, encompassing navigation and obstacle avoidance. However, traditional flocking algorithms typically rely on rigid rules and exhibit limited adaptability to diverse environments. Reinforcement learning (RL) effectively addresses this issue as a flexible and model-free framework. In this article, RL techniques are utilized to achieve navigation and obstacle avoidance for a swarm of AAVs. To enhance training efficiency, we propose an improved algorithm called heuristic guides TD3 (HGTD3) by integrating heuristic guides with the twin delayed deep deterministic policy gradient (TD3), aiming to address the protracted learning periods commonly observed in traditional RL methods. Considering the network-constrained environment, we propose the negative interference flocking algorithm (NIFA): the network interference flocking algorithm and an AAV flocking algorithm designed based on the sparrow search algorithm. NIFA can guide the losing AAV to follow the swarm and at the same time maintain the overall navigation and avoidance efficiency. Finally, we demonstrate the scalability and adaptability of HGTD3-NIFA in a simulation experiment in terms of multi-AAVs flocking and navigation.

Optical communication alignment between an unmanned aerial vehicle (UAV) and an autonomous underwater vehicle (AUV) is crucial for achieving cross sea-air optical communication. However, the high directivity of laser beams and sea surface fluctuations pose significant challenges to precise alignment. To address this issue, this paper first establishes a dynamic sea surface model and a cross sea-air optical communication channel model, utilizing the beam splitting method to determine the light spot center position on the UAV receiving plane. To achieve rapid and stable alignment, we propose an adaptive attention-based multi-agent deep reinforcement learning (MADRL) strategy with Koch Snowflake cluster structure for cross sea-air optical communication alignment (A2M-KS). This algorithm dynamically optimizes the evaluation network through an adaptive learning rate and employs an attention mechanism to enable agents to selectively focus on relevant information, thereby achieving rapid convergence and higher accuracy. Furthermore, by designing an attractive reward function, the UAV can rapidly and accurately track the light spot center, completing the alignment task with the AUV. Experimental results demonstrate that the A2M-KS algorithm outperforms baseline methods in terms of communication reliability, convergence speed, and accuracy.

Deep Reinforcement Learning (DRL) has garnered significant attention for its superior adaptability and continuous learning capabilities in the decision control of autonomous driving. However, in complex human-machine hybrid scenarios with high-dimensional continuous state spaces, such as unsignalized intersections, DRL-based autonomous driving decision-making control algorithms exhibit low sample efficiency and struggle to converge. Moreover, designing a reward function that accurately captures the complexities of human-machine interactions is highly challenging; inappropriate reward design can lead to unintended behaviors by the agent, thereby threatening system safety. To address these issues, this study proposes a Contrastive Expert-Guided Maximum Entropy Reinforcement Learning (CE-MERL) framework. This framework trains a discriminator model using both human expert datasets and the interaction experiences of the agent with the environment to identify expert policies. It then constrains the agent’s policy alignment with the expert policy as a cost within the maximum entropy reinforcement learning framework. This not only enables the agent to conduct reasonable and efficient exploration, thereby improving sample efficiency, but also allows it to learn safer, human-like policies even when using only sparse rewards. To ensure that the discriminator can continuously and effectively identify expert policies, we adopt the Vision Transformer (ViT) model as a feature encoder and introduce auxiliary tasks of contrastive learning. This method achieves efficient coupling of state and action features and mitigates overfitting caused by limited expert demonstration data. Simulation results show that in two tasks: unprotected left turn and ramp merging, our proposed CE-MERL method, without the use of shaped rewards, exhibits higher task success rates and sample efficiency compared to baseline algorithms.

Multi-UAV autonomous cooperative air warfare is an important mode of future intelligent air warfare. However, due to the complexity and uncertainty of air combat situation information, how to accurately interpret the enemy’s sustained combat intent remains a major challenge. To address this problem, we propose a Context-Aware Adaptive Feature Fusion (CAAFF) method, which can effectively utilize the time-series data of battlefield situation for hierarchical feature fusion. Specifically, the input data is first subjected to dimensionality reduction processing and feature extraction by an encoder-decoder to provide high-quality low-dimensional feature representations for further feature fusion. Next, the middle layer captures attitude changes by aggregating information from neighboring nodes via a graph attention convolutional network (GACN), flexibly fusing the features of each node, and identifying complex relationships between nodes. Finally, the mechanism of stabilizing multi-attention with self-attention is used to integrate global information at the upper layer to construct an overall feature representation of the mission and realize local-to-global posture analysis. In order to enhance the interpretation of persistent operational intent, we utilize the context-aware module to construct contextual feature representations by combining current and historical state information, thus improving the depth and interpretability of mission understanding. Finally, we combine the CAAFF method with reinforcement learning and verify its performance through multiple experiments, demonstrating the applicability and effectiveness of the method in Multi-UAV cooperative air combat.

With the proliferation of machine learning models in diverse applications, the issue of model security has increasingly become a focal point. Model steal attacks can cause significant financial losses to model owners and potentially threaten the security of their application scenarios. Traditional model steal attacks are primarily directed at soft-label black boxes, but their effectiveness significantly diminishes or even fails in hard-label scenarios. To address this, for hard-label black boxes, this study proposes an active learning-based Fusion Augmentation Model Stealing Framework (AugSteal). This framework initially utilizes large-scale irrelevant public datasets for deep filtering and feature extraction to generate reliable, diverse, and representative high-quality data subsets as the stealing dataset. Subsequently, we developed an adaptive active learning selection strategy that selects data samples with significant information gain for different black-box models, enhancing the attack’s specificity and effectiveness. Finally, to further address the trade-off between query budget and steal precision, this paper designed a Fusion Augmentation training method constituted of two different loss functions, enabling the substitute model to closely approximate the decision distribution of the target black box.The comprehensive experimental results indicate that, compared to the current state-of-the-art attack methods, our approach achieved a maximum performance gain of 8.21% in functional similarity for the substitute models in simulated black-box scenarios CIFAR10, SVHN, CALTECH256, and the real-world application Tencent Cloud API.

With the further development of Unmanned Aerial Vehicle (UAV) technologies, research on multi-UAV formations have also received more attention. Unmanned Aerial Vehicles (UAVs) cooperate with each other to form a formation group, which can give full play to the advantages that a single UAV does not have, and more capable of working in multi-task scenarios. Based on the Sierpinski fractal structure, this paper proposes a hybrid formation control architecture. To address the aggregation problem of UAV swarms, a multi-agent deep reinforcement learning (MADRL) method is used for aggregation control. To improve the efficiency, accuracy, effectiveness and scalability of MADRL in environments with a large number of UAVs, a multi-intelligent deep deterministic policy gradient method based on attention mechanism and adaptive accuracy (3A-MADDPG) is proposed. The method enables each agent to selectively focus on the information of other agents, with adaptive learning rate to dynamically learn its own critic network. The algorithm has a large learning rate in the early stage and converges quickly. The later stage of the algorithm has small learning rate and high accuracy, and combined with the fixed variable hybrid reward function designed in the paper, so the UAV can fly to the center of the expected region more accurately, making the whole cluster more accurate. Experiments show that the 3A algorithm proposed in the paper converges faster and achieves higher accuracy than the benchmark algorithm, whether it is the cluster aggregation formation or the separation and aggregation of sub-formations.

FANET (Fly-Adhoc-Network) does not rely on pre-built infrastructure, and can form a temporary network through wireless links anytime and anywhere, which has been widely used in emergency communication and disaster relief. In order to solve the problem of signal transmission fading caused by heavy fog, heavy smoke and other harsh environments, and data packet transmission failure caused by high speed movement of unmanned aerial vehicle (UAV), a cooperative transmission mode is adopted to select a relay at the data link layer for packet forwarding, which effectively improves the network communication performance. Aiming at the problem of limited communication of UAV swarm in harsh environment, a cooperative medium access control protocol named energy consumption and channel gain cooperative medium access control protocol (EC-CMAC) for FANET is proposed. The protocol estimate the transmit power according to the channel propagation model, and further propose a relay selection strategy according to the estimate transmit power, the residual energy of the node, the direction and position of the node, and adaptively selecting the transmission mode. During cooperative transmission, it selected one-hop neighbors to forward the message, and extended the network lifetime by optimizing the energy consumption of the UAVs. The protocol is verified both statically and dynamically in MATLAB environment. The evaluation results of end-to-end delay, network throughput, network lifetime and packet transmission ratio testify that EC-CMAC can achieve longer network lifetime and higher packet delivery rate with little delay and throughput loss in harsh transmission environment and high speed movement of UAVs.

Rescue assignments strategy are crucial for multiple Autonomous Underwater Vehicle (multi-AUV) systems in three dimensional (3-D) complex underwater environments. Considering the requirements of rescue missions, multi-AUV systems need to be cost-effective, fast-rescuing, and less concerned about the relationship between rescue missions. The real-time rescue plays a vital role in the multi-AUV system with the characteristics mentioned above. In this paper, we propose an efficient Reward acting on Reinforcement Learning and Particle Swarm Optimization (R-RLPSO), to provide a strategy of real-time rescue assignment for the multi-AUV system in the 3-D underwater environment. This strategy consists of the following three parts. Firstly, we present a reward-based real-time rescue assignment algorithm. Secondly, we propose an Attraction Rescue Area containing a Rescue Area. For the waypoints in each Attraction Rescue Area, the reward is calculated by a linear reward function. Thirdly, to speed up the convergence of the R-RLPSO and mark the rescue states of Attraction Rescue Area and rescue area, we develop a Reward Coefficient based on the reward of all Attraction Rescue Areas and Rescue Areas. Finally, simulation results show that the system based on R-RLPSO is more cost-effective and time-saving than that of based on comparison algorithms ISOM and IACO.

The recognition and classification of the remote sensing image is a key technology in the application of remote sensing image, which has been one of the research hotspots. Extreme learning machine (EML), is a kind of machine learning method, which has advantages of fast learning speed and good generalization capability, and has been getting more and more attention from the researchers. Based on the two research hotspots, in this paper, a method for recognition of fold structure information in the remote sensing image based on ELM is proposed. Through the comparison of results obtained by this method and artificial interpretation, it shows that the method has good accuracy and strong applicability, and also, has important theoretical basis and practical application value in guiding the planning and construction and operation maintenance of special regional highway.