In this paper, an enhanced deep reinforcement learning approach is presented for unmanned aerial vehicles (UAVs) operating in dynamic and potentially hazardous environments. Initially, the capability to discern obstacles from visual data is achieved through the application of the Yolov8-StrongSort technique. Concurrently, a novel data storage system for deep Q-networks (DQN), named dynamic data memory (DDM), is introduced to hasten the learning process and convergence for UAVs. Furthermore, addressing the issue of UAVs’ paths veering too close to obstacles, a novel strategy employing an artificial potential field to adjust the reward function is introduced, which effectively guides the UAVs away from proximate obstacles. Rigorous simulation tests in an AirSim-based environment confirm the effectiveness of these methods. Compared to DQN, dueling DQN, M-DQN, improved Q-learning, DDM-DQN, EPF (enhanced potential field), APF-DQN, and L1-MBRL, our algorithm achieves the highest success rate of 77.67%, while also having the lowest average number of moving steps. Additionally, we conducted obstacle avoidance experiments with UAVs with different densities of obstacles. These tests highlight fast learning convergence and real-time obstacle detection and avoidance, ensuring successful achievement of the target.

Unmanned aerial vehicles (UAVs) have made significant advances in autonomous sensing, particularly in the field of precision agriculture. Effective path planning is critical for autonomous navigation in large orchards to ensure that UAVs are able to recognize the optimal route between the start and end points. When UAVs perform tasks such as crop protection, monitoring, and data collection in orchard environments, they must be able to adapt to dynamic conditions. To address these challenges, this study proposes an enhanced Q-learning algorithm designed to optimize UAV path planning by combining static and dynamic obstacle avoidance features. A shortest distance priority (SDP) strategy is integrated into the learning process to minimize the distance the UAV must travel to reach the target. In addition, the root mean square propagation (RMSP) method is used to dynamically adjust the learning rate according to gradient changes, which accelerates the learning process and improves path planning efficiency. In this study, firstly, the proposed method was compared with state-of-the-art path planning techniques (including A-star, Dijkstra, and traditional Q-learning) in terms of learning time and path length through a grid-based 2D simulation environment. The results showed that the proposed method significantly improved performance compared to existing methods. In addition, 3D simulation experiments were conducted in the AirSim virtual environment. Due to the complexity of the 3D state, a deep neural network was used to calculate the Q-value based on the proposed algorithm. The results indicate that the proposed method can achieve the shortest path planning and obstacle avoidance operations in an orchard 3D simulation environment. Therefore, drones equipped with this algorithm are expected to make outstanding contributions to the development of precision agriculture through intelligent navigation and obstacle avoidance.

Unmanned Aerial Vehicle (UAV) path planning is critical for ensuring flight safety and enhancing mission execution efficiency. This problem is typically formulated as a complex, multi-constrained, and nonlinear optimization task, often addressed using meta-heuristic algorithms. The Crested Porcupine Optimizer (CPO) has become an excellent method to solve this problem; however, the standard CPO has limitations, such as the lack of adaptive parameter tuning to adapt to complex environments, slow convergence, and the tendency to fall into local optimal solutions. To address these issues, this paper proposes an algorithm named QCPO, which integrates CPO with Q-learning to improve UAV path optimization performance. Q-learning is employed to adaptively adjust the key parameters of the CPO, thereby overcoming the limitations of traditional fixed-parameter settings. Inspired by the porcupine’s defense mechanisms, a novel audiovisual coordination strategy is introduced to balance visual and auditory responses, accelerating convergence in the early optimization stages. A refined position update mechanism is designed to prevent excessive step sizes and boundary violations, enhancing the algorithm’s global search capability. A B-spline-based trajectory smoothing method is also incorporated to improve the feasibility and smoothness of the planned paths. In this paper, we compare QCPO with four outstanding heuristics, and QCPO achieves the lowest path cost in all three test scenarios, with path cost reductions of 30.23%, 26.41%, and 33.47%, respectively, compared to standard CPO. The experimental results confirm that QCPO offers an efficient and safe solution for UAV path planning.

The acquisition of plant phenotypic traits is essential for selecting superior varieties, improving crop yield, and supporting precision agriculture and agricultural decision-making. Therefore, it plays a significant role in modern agriculture and plant science research. Traditional manual measurements of phenotypic traits are labor-intensive and inefficient. In contrast, combining 3D reconstruction technologies with autonomous vehicles enables more intuitive and efficient trait acquisition. This study proposes a 3D semantic reconstruction system based on an improved ORB-SLAM3 framework, which is mounted on an unmanned vehicle to acquire phenotypic traits in tomato cultivation scenarios. The vehicle is also equipped with the A * algorithm for autonomous navigation. To enhance the semantic representation of the point cloud map, we integrate the BiSeNetV2 network into the ORB-SLAM3 system as a semantic segmentation module. Furthermore, a two-stage filtering strategy is employed to remove outliers and improve the map accuracy, and OctoMap is adopted to store the point cloud data, significantly reducing the memory consumption. A spherical fitting method is applied to estimate the number of tomato fruits. The experimental results demonstrate that BiSeNetV2 achieves a mean intersection over union (mIoU) of 95.37% and a frame rate of 61.98 FPS on the tomato dataset, enabling real-time segmentation. The use of OctoMap reduces the memory consumption by an average of 96.70%. The relative errors when predicting the plant height, canopy width, and volume are 3.86%, 14.34%, and 27.14%, respectively, while the errors concerning the fruit count and fruit volume are 14.36% and 14.25%. Localization experiments on a field dataset show that the proposed system achieves a mean absolute trajectory error (mATE) of 0.16 m and a root mean square error (RMSE) of 0.21 m, indicating high localization accuracy. Therefore, the proposed system can accurately acquire the phenotypic traits of tomato plants, providing data support for precision agriculture and agricultural decision-making.

It is a challenging and meaningful task to carry out drone-based livestock monitoring in high-altitude and cold regions. The purpose of AI is to execute automated tasks and to solve practical problems in actual applications by combining the software technology with the hardware carrier to create integrated advanced devices. Only in this way, the maximum value of AI could be realized. In this paper, a real-time tracking system with dynamic target tracking ability is proposed. It is developed based on the tracking-by-detection architecture using YOLOv7 and DeepSORT algorithms for target detection and tracking, respectively. To address the existing problems of the DeepSORT algorithm, the following two optimizations are made: (1) Optical flow is used to compensate the Kalman filter for improvement of the prediction accuracy; (2) A low-confidence trajectory filtering method is adopted to reduce the influence of unreliable detection on target tracking. In addition, an visual servo controller for the UAV is designed to enable the automated tracking task. Finally, the system is tested using the Tibetan yaks living in the Tibetan Plateau as the tracking targets, and the results reveal the real-time multiple tracking ability and the ideal visual servo effect of the proposed system.

Maize plant height and stem diameter can effectively reflect plant growth status and lodging resistance. Therefore, accurate measurement of these parameters enables precise monitoring of plant growth and provides data to support agricultural decision-making. LiDAR enables the acquisition of three-dimensional (3D) crop point clouds, providing a direct source of data for phenotypic parameter extraction. However, existing 3D point cloud acquisition systems for crop phenotyping are often costly and offer limited operational flexibility under field conditions. This study proposes a workflow for extracting key phenotypic parameters from individual field-grown maize plants by integrating 3D reconstruction with point cloud instance segmentation. First, field-grown maize plants are reconstructed in 3D using a LiDAR–SLAM system mounted on an unmanned ground vehicle (UGV). Because accurate segmentation of individual maize plants from field LiDAR point clouds is challenging, a density-adaptive strategy is incorporated into an instance segmentation network based on central-axis offset aggregation. Specifically, each point is projected toward the central axis of its corresponding plant, and local point density is used to guide individual-plant instance segmentation. Plant height and stem diameter are subsequently extracted automatically from the segmented point cloud of each plant and evaluated against manual measurements. Experimental results show that the density-adaptive instance segmentation method achieves an mAP50-95 of 77.84%, representing an improvement of 8.54%. The mean absolute errors (MAEs) are 7.52 cm for plant height and 1.02 mm for stem diameter. These results demonstrate that, under the tested grain-filling-stage field plots, the proposed method enables effective individual-plant segmentation and extraction of key phenotypic parameters. Given that validation was limited to two plots acquired on a single date, claims of broad field robustness should be interpreted within this experimental scope.