Medical image recognition methods based on deep learning typically have extremely high requirements for the distribution consistency of source domain data and target domain data. In real-world deployment scenarios, there are often significant differences in target domain data generated by devices from different medical centers and manufacturers. When a medical image recognition model recognizes target domain data that is different from its source domain data distribution, its generalization performance will decrease, greatly reducing the credibility of the medical image recognition model in practical applications.In response to the above issues, this article proposes a parameter adaptive tuning algorithm for medical image recognition model testing. The algorithm uses parallel networks to learn the characteristics and distribution of target domain data, while keeping the parameters of the main network unchanged during adaptation. By optimizing the parameters of the parallel network through backpropagation and gradient descent, it can avoid catastrophic forgetting of the model while also learning the characteristics and distribution of target domain data to a certain extent. In the prediction stage, the fusion network is used to predict the target and improve the generalization performance of the model. This article uses YOLOv5 as the basic model and proposes a parameter adaptive tuning algorithm for medical image recognition model testing. The accuracy of the algorithm on datasets FPPD and DRCOLO has been improved to 94.2% and 97.8%, and the mAP on multiple Disease-Colo video datasets has been improved to 30.5% and 41.7%, verifying the effectiveness of the algorithm.

As an acute disease, facial paralysis has high requirements for the point of treatment, and if we cannot accurately grasp the period in which the patient is suffering from the disease, we may miss the optimal time for treatment, which may affect the condition. We need to establish intelligent auxiliary diagnosis methods for facial palsy, in which AI-based methods rely on a large number of training samples. However, facial paralysis image samples are small, resulting in low model accuracy, which cannot meet the needs of clinical applications. Therefore, we plan to expand the facial palsy samples through data augmentation methods to reduce the training sample requirements and improve the accuracy of facial palsy grading. We propose the method of fusing generative adversarial networks, where the collected facial paralysis samples are trained by DragGAN model to expand the sample space, and then input into ResNet152 model for classification training. The classification effect of facial palsy images is effectively improved by the principle of data augmentation before classification. The accuracy of this experimental method on the MEEI facial palsy dataset is 85.5%, the recall is 81.4%, and the F1-score is 80.7%, which are 13.1%, 16.7%, and 15.8%, respectively compared to the method without data augmentation. A facial palsy classification model incorporating generative adversarial networks achieves higher classification accuracy compared to methods without data augmentation, providing a new approach to facial palsy classification.

The efficiency of informed path planning algorithms is contingent upon how quickly the planner can find the initial solution and the associated overhead involved in collision detection. Existing informed planners do not fully exploit the information contained in historical collision detection results, resulting in additional unnecessary collision detections. Furthermore, they optimize paths through rewiring before discovering an initial solution, which not only hampers the planner’s space exploration, but also generates a superfluous amount of unproductive over-head. To address the shortcomings of existing algorithms, this paper proposes an Obstacle-Sensitive and Initial-Solution-first path planning algorithm (OSIS). OSIS uses historical collision detection results to predict the distribution of obstacles in space and utilizes an initial-solution-first path optimization strategy to avoid useless path optimization. Experiments show that OSIS can efficiently bypass obstacles and converge the cost of the solution compared to existing algorithms.

In recent years, Unmanned Aerial Vehicle (UAV) networks are widely used in both military and civilian scenarios. However, due to the distributed nature, they are also vulnerable to threats from adversaries. Time delay attack is a type of internal attack which maliciously delays the transmission of data packets and further causes great damage to UAV networks. Furthermore, it is easy to implement and difficult to detect due to the avoidance of packet modification and the unique characteristics of UAV networks. However, to the best of our knowledge, there is no research on time delay attack detection in UAV networks. In this paper, we propose an Efficient Time Delay Attack Detection Framework (ETD). First, we collect and select delay-related features from four different dimensions, namely delay, node, message and connection. Meanwhile, we utilize the pre-planned trajectory information to accurately calculate the real forwarding delay of nodes. Then, one-class classification is used to train the detection model, and the forwarding behaviors of all nodes can be evaluated, based on which their trust values can be obtained. Finally, the K-Means clustering method is used to distinguish malicious nodes from benign ones according to their trust values. Through extensive simulation, we demonstrate that ETD can achieve higher than 80% detection accuracy with less than 2.5% extra overhead in various settings of UAV networks and different routing protocols.

With the development of airborne equipment and integrated avionics technology, the unmanned aerial vehicle (UAV) network replaces human beings in many works. However, the limited energy capacity of UAVs has great restrictions on long-time missions. To prolong the lifetime of the UAV network, we introduce the wireless static chargers (WSCs) into the UAV network, and a nondisruptive wireless rechargeable UAV network (WRUN) model is proposed, in which UAVs can be wirelessly charged without returning back to the charging platform and WSCs are scheduled to turn on and release energy only in the charging time periods. The goal of this article is to minimize the energy waste of WSCs under the premise that UAVs will not run out of energy. To calculate the efficient charging time periods of WSCs, we first discretize flight paths of UAVs so that the nondisruptive charging time schedule problem (nCTSP) that has an infinite solution space both in the spatial and time dimensions can be formalized as an optimization problem. Then, we propose the baseline algorithm exhaust candidate solutions (ECSs) to calculate the charging time periods and propose an improved algorithm using the idea of pruning (PECS) to reduce the computational complexity of ECS. Finally, experiments are conducted and PECS achieves better performance in terms of saving energy consumption and maximizing energy utilization.

In order to stablish and evaluate the drug-disease network, we take the common thyroid diseases in daily life as an example to study the modeling and evaluation of drug-disease network. Firstly, mining the highest-frequency drug in the medical records of thyroid tumors of overseas medical masters; Then, find out the thyroid tumor target and drug target in the disease target database and the traditional Chinese medicine pharmacology analysis platform, and compare and find out the potential active target of the drug on thyroid tumor. After signal pathway analysis, chemical composition target network and target signal pathway network model diagrams were constructed. The experimental results show that there are active components in the drug pair, which act on thyroid tumors through multiple signal pathways. The establishment and evaluation of drug disease network can provide a theoretical basis for the inheritance of clinical experience of famous traditional Chinese medicine.