Deep Q-network(DQN) is one of the standard methods to solve the Unmanned Aerial Vehicle(UAV) path planning problem. However, the way agent deepens its cognition of the environment through frequent random trial-and-error leads to slow convergence. This paper proposes an optimized DQN with Artificial Potential Field(APF) as prior knowledge called B-APFDQN for path planning. Replacing the traditional neural network which has only one Q-value output with a multi-output neural network to promote the training process in combination with APF. Furthermore, a SA-ε-greedy algorithm that can automatically adjust the stochastic exploration frequency with steps and successes is proposed in order to prevent the agent from falling into local optimum. We remove the nodes that do not affect the path connectivity and apply the B-spline algorithm to make the path shorter and smoother. Simulation experiments show that the proposed B-APFDQN algorithm performs better than the classical DQN, has a strong ability to avoid falling into local optimum, and the obtained paths are smooth and shorter, which proves the effectiveness of B-APFDQN in the UAV path planning problem.
Neural networks have been widely employed in the field of object detection. Transformers enable effective object detection through global context awareness, modular design, scalability, and adaptability to diverse target scales. However, small object detection requires careful consideration due to its comprehensive computations, data requirements, and real-time performance challenges. To address these issues, we present SwinVision , an innovative framework for small object detection in low-light environments. This research shows a balanced approach between computational efficiency and detection accuracy for advancing object detection in low-light scenarios. Firstly, a Swin Transformer-based computing network is introduced and optimized for object detection in large-scale areas. The framework balances computational power and resource efficiency, surpassing conventional transformers. Secondly, we present the STLE module, which enhances the features of low-light images for beneficial object detection. The last building block is a specialized Swin-based detection block for accurate detection of small, detailed objects in resource-constrained scenarios. Experiments conducted on the VisDrone dataset significantly ameliorated existing methods such as YOLOv8x, with a 6.31% increase in mAP and 12.55% in AP50. SwinVision’s effectiveness in low-light environments, especially with small objects, establishes a foundation for robust detection systems adapting to various environmental challenges.
Unmanned aerial vehicles (UAVs) have increasingly become indispensable instruments for surveillance, reconnaissance, and various missions due to their advanced capabilities. In intricate and dynamic operational contexts, especially during ultra-low altitude missions, inadequate communication and coordination impede UAVs’ capacity to avoid other UAVs and moving objects. In this instance, attaining accurate path planning for UAVs constitutes a significant technological problem that requires immediate attention. This paper introduces an innovative algorithm, ST-D3QN, which leverages deep reinforcement learning (DRL) to optimize UAV trajectories based on optimizing the Dueling Double Deep Q-Network (D3QN) and A* (A-star) algorithm. Our approach refines the optimization of sparse rewards by toggling between the two methods, thereby maximizing the long-term benefits for UAVs. The greedy strategy has been refined to prevent the algorithm from becoming ensnared in local optima. Our proposed method is benchmarked against state-of-the-art path planning algorithms through meticulously designed simulation tests. The experimental results indicate that the ST-D3QN algorithm achieves the fastest convergence rate during training and excels in optimizing sparse rewards. It also significantly reduces the risk of collisions and enhances the success rates of UAV path planning, establishing itself as an efficient and effective solution for determining the shortest path for UAVs in complex environments.
1 January 2023 in Computer Modeling in Engineering & Sciences
Unmanned Aerial Vehicles (UAVs) are widely used and meet many demands in military and civilian fields. With the continuous enrichment and extensive expansion of application scenarios, the safety of UAVs is constantly being challenged. To address this challenge, we propose algorithms to detect anomalous data collected from drones to improve drone safety. We deployed a one-class kernel extreme learning machine (OCKELM) to detect anomalies in drone data. By default, OCKELM uses the radial basis (RBF) kernel function as the kernel function of the model. To improve the performance of OCKELM, we choose a Triangular Global Alignment Kernel (TGAK) instead of an RBF Kernel and introduce the Fast Independent Component Analysis (FastICA) algorithm to reconstruct UAV data. Based on the above improvements, we create a novel anomaly detection strategy FastICA-TGAK-OCELM. The method is finally validated on the UCI dataset and detected on the Aeronautical Laboratory Failures and Anomalies (ALFA) dataset. The experimental results show that compared with other methods, the accuracy of this method is improved by more than 30%, and point anomalies are effectively detected.
1 January 2019 in Computer Modeling in Engineering & Sciences
Path planning remains a focal point in Unmanned Aerial Vehicle (UAV) research, with autonomous path planning in unknown environments emerging as a particularly active area. Deep Reinforcement Learning (DRL) has become a standard approach for addressing path planning challenges in such environments. However, traditional DRL algorithms often suffer from slow convergence. To address this issue, this paper introduces an enhanced deep reinforcement learning algorithm, termed Firefly Algorithm-enhanced Deep Q-Network (FADQN). In FADQN, the action output from the Firefly Algorithm is integrated into the Deep Q-Network to guide early action selection, thereby accelerating convergence. To balance exploration and exploitation, an adaptive ε-greedy strategy is proposed. Furthermore, a prioritized experience replay mechanism is employed to increase the likelihood of the agent learning from critical experiences. Simulation experiments demonstrate that FADQN achieves shorter paths (reducing the path length by 7.32% in complex environments), fewer turns, faster and more stable convergence, and higher success rates across various environments compared to traditional DQN. These results validate the effectiveness of FADQN in solving UAV path planning problems.