As a respiratory infection, pneumonia has gained great attention from countries all over the world for its strong spreading and relatively high mortality. For pneumonia, early detection and treatment will reduce its mortality rate significantly. Currently, X-ray diagnosis is recognized as a relatively effective method. The visual analysis of a patient’s X-ray chest radiograph by an experienced doctor takes about 5 to 15 minutes. When cases are concentrated, this will undoubtedly put tremendous pressure on the doctor’s clinical diagnosis. Therefore, relying on the naked eye of the imaging doctor has very low efficiency. Hence, the use of artificial intelligence for clinical image diagnosis of pneumonia is a necessary thing. In addition, artificial intelligence recognition is very fast, and the convolutional neural networks (CNNs) have achieved better performance than human beings in terms of image identification. Therefore, we used the dataset which has chest X-ray images for classification made available by Kaggle with a total of 5216 train and 624 test images, with 2 classes as normal and pneumonia. We performed studies using five mainstream network algorithms to classify these diseases in the dataset and compared the results, from which we improved MobileNet’s network structure and achieved a higher accuracy rate than other methods. Furthermore, the improved MobileNet’s network could also extend to other areas for application.

Marine unmanned vehicle is a novel robot widely used in ocean observation, and its accurate control is of significance to their path planning. We want to find a method to predict the velocity and course of this robot, which can help us realize the accurate control of it. The paper proposed a promising type of hybrid robotic fish (HRF), which can realize two kinds of motion modes on the sea surface. Firstly, the configuration and dynamic model of the HRF were analyzed elaborately. Then, to realize accurate velocity prediction under two kinds of motion modes of HRF, the influence factors are presented in a complex marine environment. Based on the influence factors to its maneuverability, such as wind or wave parameters, a velocity prediction algorithm based on back-propagation neural network (BPNN) was introduced. However, BPNN has the disadvantages of extended learning and training time, easily falling into local optimum. Then we found that genetic algorithm (GA), which is a kind of evolutionary algorithm, is suitable for our problem. Therefore, the accuracy and efficiency of the prediction algorithm were improved by adopting the genetic algorithm to optimize the weight and threshold of BPNN. Taking the experimental data from the pool test, the back-propagation neural network with genetic algorithm (GA-BPNN) forecasting model was established. Besides, the other prediction methods were compared and evaluated under the same assessment criterion to validate the proposed forecasting model. The experimental results demonstrate that the GA-BPNN model has higher accuracy and efficiency compared with other prediction algorithms, which verifies the feasibility of the velocity prediction model for hybrid robotic fish in complex ocean environments.

The underwater glider (UG) is a new type of long-endurance marine vehicle that is currently widely used in marine environment surveillance. However, the limited capability of underwater communication and the uncertainty of the marine environment leads to a large deviation between the dead-reckoned trajectory and the real trajectory of UG. Therefore, a localization and navigation method based on ocean current data is proposed in this study. Firstly, the working principle and configuration of UG are illustrated, and the dynamic model is also analyzed. Then, the depth-averaged current is derived with motion parameters of UG, and the ocean current data from the Hybrid Coordinate Ocean Model is also introduced and processed. Meanwhile, to make an accurate localization, the DAC is stratified according to the layer weight extracted from the processed ocean current model to obtain layered DAC (LDAC). Subsequently, because the sample intervals of sensors are asynchronous with LDAC, a new matching algorithm based on sample time difference is presented to match the information of sensors and LDAC. Furthermore, the matched data is applied to localization for accurate dead reckoning. Finally, based on the data collected from two experiments, the LDAC-aided localization method proposed in this paper is compared with other benchmark positioning methods, including dynamic model-aided localization, sea current-aided localization, and acoustic beacon-aided localization methods. The results indicate that the proposed new method achieves better localization performance than other benchmark methods. In addition, the proposed method could be applied to other autonomous vehicles with communication constraints.

This article proposes a novel method for generating the regular virtual tube for safe task execution by UAV swarms in fire dynamic propagation scenarios with static obstacles. Cellular automata is utilized to simulate the dynamic propagation characteristics of fire. We employ the empirical Wang Zhengfei model, taking into account environmental vegetation characteristics, initial wind speed, and wind correction coefficient, to define the state transition function for the cellular automaton. The generation framework for regular virtual tube is divided into two parts: front-end search and back-end optimization. In the front-end, a collision-free trajectory is computed using a search-based approach. In the back-end, a reference curve generation method with soft constraints is employed, and it is solved using gradient information. Subsequently, an optimization method for the radius of the regular virtual tube is provided based on the reference curve. Simulation results demonstrate that the real-time generated regular virtual tube can adapt to dynamic fire scenarios and are more robust in complex obstacle scenarios compared to existing state-of-the-art approaches.

Underwater gliders (UGs) represent a class of autonomous underwater vehicles renowned for their extended operational endurance, capable of traversing thousands of kilometers. As the primary environmental factor influencing UG navigation, ocean currents profoundly affect path planning strategies, with temporal variability exhibiting significant regional disparities across marine domains. This study introduces a hybrid path-planning methodology for transoceanic UGs, synergizing classical algorithms with reinforcement learning techniques. The framework initiates by constructing a transoceanic current model through integration of multitemporal hydrodynamic data from the Global Ocean Physics Analysis and Forecast system. Subsequently, the Dijkstra algorithm generates initial trajectory planning within a static current field derived from long-term averaged flow patterns. Leveraging these preliminary waypoints, a Q-learning algorithm performs segmented trajectory optimization using real-time current data. The refined path undergoes final processing through a smoothing algorithm to yield navigable routes compatible with UG operational constraints. Accompanying this methodology, a dedicated software platform, UG path planning platform (UGPPP), facilitates path planning and performance evaluation. Validation through case studies in the South China Sea, Western Pacific, and transoceanic regions demonstrates the proposed method’s superior overall performance compared to benchmark approaches, achieving a 35.946% hydrodynamic energy utilization rate. This systematic approach establishes a foundational framework for optimizing long-range marine vehicle navigation in dynamic oceanic environments.