Despite the ability of semi-supervised medical image segmentation to achieve superior results with limited labeled data and extensive unlabeled data, prevailing methods encounter a significant challenge: the detrimental cycle of pseudo-label noise resulting in model bias and error accumulation, which ultimately degrades performance. In this paper, we propose a differential teacher dynamic co-evolution mechanism: initially, an uncertainty perception-driven confidence fusion module (UPC) is developed, converting the uncertainty region into learning opportunities to maintain essential anatomical structures. The Dynamic Update Mechanism (DUM) is concurrently proposed to adaptively select and enhance the teacher models utilizing the pixel-level confidence graph to mitigate noise propagation. Furthermore, we propose multi-view collaborative distillation optimization (MCD), which facilitates bidirectional knowledge transfer between teachers and students via multi-level knowledge distillation, and addresses mistake accumulation to transcend the noise-induced local decision-making boundary. Experiments on the LA and ACDC datasets demonstrate that our method markedly outperforms current semi-supervised segmentation methods.

With the growing diversity and density of mobile nodes, indoor wireless channels are becoming increasingly complex. Existing channel modeling methods struggle to balance accuracy and adaptability, which calls for low-complexity models capable of capturing dynamic indoor environments efficiently. This paper proposes a novel propagation graph (PG) framework that models the channel effects of dynamic objects indoors by designing a scattering-aware point cloud neural network (SPointNet). The proposed PG framework explicitly incorporates reflection, transmission, and diffuse scattering into channel modeling, and employs a physics-aware scatterer discretization and classification strategy, which reduces the complexity of the conducted graph. Then, SPointNet enables fast estimation of scattering coefficients, allowing the model to bypass exhaustive analysis of material properties. Finally, we conduct channel measurements in real indoor environments to validate the proposed approach. Experimental results show that the proposed model accurately models channel responses while significantly reducing modeling time compared to traditional PG-based methods.

The deployment of object detection tasks on embedded or mobile platforms has become increasingly prevalent, driven by the heightened demand across various scenarios. However, for object recognition tasks such as gesture recognition, the use of overly complex network models presents a formidable obstacle in achieving real-time detection tasks, and the majority of lightweight convolution menthods based on depth-separated convolution lack accuracy. In this paper, we propose a lightweight and highly accurate convolutional neural network for gesture recognition on terminal device (GRTD-Net), specially designed for devices with scarce computing power and tight hardware resources. In GRTD-Net, we proposed the convolution method R2SGConv that masterfully harmonizes model size and accuracy, elegantly achieving a delicate balance between efficiency and lightweight design. Moreover, we propose a neck network paradigm with good feature fusion capability to compensate for the accuracy degradation due to the use of lightweight convolutional modules in neck networks. Experimental results show that the proposed GRTD-Net model improves the mAP0.5and mAP0.95by 0.8% and 2.2%, reduces model parameters by 35.2%, increases FPS by 20.7%, and reduces the inference latency by 2.9 ms, compared with the popular YOLOv5 algorithm on the dataset Gesture. We successfully deployed GRTD-Net on ARM devices and proved its practicality in constrained environments.

Internet of Underwater Things (IoUT) is applied to ocean research by connecting underwater sensing devices. It has ability to maintain efficient communication even in challenging environments and with limited power resources. However, it hinders the efficient storage and forwarding of huge amounts of underwater data due to the traditional IP architecture. As a future network architecture, underwater named data networking (UNDN) is considered an effective architecture of IoUT. While naive UNDN cannot support active forwarding when dangers occur in underwater environments, which may cause severe consequences. In this article, we propose an event-based communication architecture called push-hybrid UNDN, which can achieve active and timely forwarding when critical events occur. To realize critical data forwarding without waiting for user requests, passively accepted interest is replaced by actively forwarded beacon during the critical events. The push-hybrid UNDN is analyzed and compared with the naive one in transmission performance to highlight the advantages of incorporating push mechanism. Energy consumption of push-hybrid UNDN, naive UNDN, and push-based UNDN is compared in different scenarios to demonstrate the advantages of hybrid design. Simulation results verify that our scheme achieves shorter transmission delay, more stable and higher packet delivery ratio, less traffics, and lower energy consumption than the existing UNDN architectures.

Internet of Underwater Things (IoUT) needs to maintain effective communication even under the circumstances of severe environments and limited energy. Named data networking (NDN), a future network architecture, is starting to be used for IoUT as an effective architecture implementation. Despite having a good performance of data transmission, Underwater named data networking (UNDN) nevertheless faces some security risks, such as Denial-of-Service (DoS) brought by interest flooding attacks (IFAs). This article proposes a novel DoS attack, Synergetic DoS (SDoS), which can cause hiding damages to router’s content store (CS), pending interest table (PIT), and forwarding information base (FIB). We not only study the basic synergetic attack model of SDoS but also analyze some possible attack variants. Simulation results illustrate that SDoS entirely invalidates the only IFA detection algorithm in UNDN. Compared to ordinary IFAs, SDoS attacks increase network traffic fourfold. Furthermore, we discover a unique infection problem in UNDN and propose a countermeasure named Trident, which has meticulously designed adaptive threshold,Double Trialfor attacker identification, and a self-proving mechanism based on leaky bucket. Experiment results demonstrate that Trident can detect and resist not only IFAs but also SDoS attacks effectively. Meanwhile, Trident also achieves good defense performance on the variants of SDoS and can take on burst traffic and network congestion robustly.

Deep learning (DL) technologies bring new threats to network security. Website fingerprinting attacks (WFA) using DL models can distinguish victim’s browsing activities protected by anonymity technologies. Unfortunately, traditional countermeasures (website fingerprinting defenses, WFD) fail to preserve privacy against DL models. In this paper, we apply adversarial example technology to implement new WFD with static analyzing (SA) and dynamic perturbation (DP) settings. Although DP setting is close to a real-world scenario, its supervisions are almost unavailable due to the uncertainty of upcoming traffics and the difficulty of dependency analysis over time. SA setting relaxes the real-time constraints in order to implement WFD under a supervised learning perspective. We propose Greedy Injection Attack (GIA), a novel adversarial method for WFD under SA setting based on zero-injection vulnerability test. Furthermore, Sniper is proposed to mitigate the computational cost by using a DL model to approximate zero-injection test. FCNSniper and RNNSniper are designed for SA and DP settings respectively. Experiments show that FCNSniper decreases classification accuracy of the state-of-the-art WFA model by 96.57% with only 2.29% bandwidth overhead. The learned knowledge can be efficiently transferred into RNNSniper. As an indirect adversarial example attack approach, FCNSniper can be well generalized to different target WFA models and datasets without suffering fatal failures from adversarial training.

Semantic segmentation of aerial images plays an important role in urban area monitoring. But the diversity of buildings makes segmentation a hard task. To detect buildings from aerial images more precisely, this paper proposes a pixel-level segmentation method, named Attention Residual U-net (ARU-net). ARU-net adds two major part into the framework of U-net, i.e. attention path and residual connection, focusing on feature reuse. Attention path utilizes attention mechanism to capture spatial feature details. Residual connection implies the semantic information flow through a 1×1 convolution similar to the residual form. ARU-net can be trained end-to-end. Experiments are conducted to evaluate the effectiveness of the proposed model on the Inria Aerial Image Labeling Dataset. Results indicate that ARU-net outperforms other baselines with an accuracy of 93.84% and intersection over union (IoU) of 60.90%.

Due to the harsh environment and energy limitation, maintaining efficient communication is crucial to the lifetime of Underwater Sensor Networks (UWSN). Named Data Networking (NDN), one of future network architectures, begins to be applied to UWSN. Although Underwater Named Data Networking (UNDN) performs well in data transmission, it still faces some security threats, such as the Denial-of-Service (DoS) attacks caused by Interest Flooding Attacks (IFAs). In this paper, we present a new type of DoS attacks, named as Synergetic Denial-of-Service (SDoS). Attackers synergize with each other, taking turns to reply to malicious interests as late as possible. SDoS attacks will damage the Pending Interest Table, Content Store, and Forwarding Information Base in routers with high concealment. Simulation results demonstrate that the SDoS attacks quadruple the increased network traffic compared with normal IFAs and the existing IFA detection algorithm in UNDN is completely invalid to SDoS attacks. In addition, we analyze the infection problem in UNDN and propose a defense method Trident based on carefully designed adaptive threshold, burst traffic detection, and attacker identification. Experiment results illustrate that Trident can effectively detect and resist both SDoS attacks and normal IFAs. Meanwhile, Trident can robustly undertake burst traffic and congestion.

In this paper, we consider a two-layer caching-aided network, where a single server consisting of a library of N files connects with multiple relays, each equipped with a cache memory of M1files and each relay connects with a distinct set of users, each equipped with a cache memory of M2files. We design a caching scheme that exploits the spared transmission time resource by constructing a concurrent transmission between the two layers. It is shown that the caching scheme is order optimal and achieves an additive parallel gain compared to the previously known caching scheme. Also, we show that for the two-relay case, if each relay's caching size M1equals to 0.382N, our scheme achieves the optimal delay as M1= N, implying that increasing the relay's cache size will not always reduce the transmission delay.

In this paper, we consider the coded-caching broadcast network with user cooperation, where a server connects with multiple users and the users can cooperate with each other through a cooperation network. We propose a centralized coded caching scheme based on a new deterministic placement strategy and a parallel delivery strategy. It is shown that the new scheme optimally allocate the communication loads on the server and users, obtaining cooperation gain and parallel gain that greatly reduces the transmission delay. Furthermore, we show that the number of users who parallelly send information should decrease when the users' caching size increases. In other words, letting more users parallelly send information could be harmful. Finally, we derive a constant multiplicative gap between the lower bound and upper bound on the transmission delay, which proves that our scheme is order optimal.