Abstract A botnet is a group of hijacked devices that conduct various cyberattacks, which is one of the most dangerous threats on the internet. Organizations or individuals use network traffic to mine botnet communication behavior features. Network traffic often contains individual users’ private information, such as website passwords, personally identifiable information, and communication content. Among the existing botnet detection methods, whether they extract deterministic traffic interaction features, use DNS traffic, or methods based on raw traffic bytes, these methods focus on the detection performance of the detection model and ignore possible privacy leaks. And most methods are combined with machine learning and deep learning technologies, which require a large amount of training data to obtain high-precision detection models. Therefore, preventing malicious persons from stealing data to infer privacy during the botnet detection process has become an issue worth pondering. Based on this problem, this article proposes a privacy-enhanced framework with deep learning for botnet detection. The goal of this framework is to learn a feature extractor. It can hide the private information that the attack model tries to infer from the intermediate anonymity features, while maximally retaining the interactive behavior features contained in the original traffic for botnet detection. We design a privacy confrontation algorithm based on a mutual information calculation mechanism. This algorithm simulates the game between the attacker trying to infer private information through the attack model and the data processor retaining the original content of the traffic to the maximum extent. In order to further ensure the privacy protection of the feature extractor during the training process, we train the feature extractor in the federated learning training mode. We extensively evaluate our approach, validating it on two public datasets and comparing it with existing methods. The results show that our method can effectively ensure detection accuracy on the basis of removing private information. For the CTU-13 dataset, the detection framework achieves the best detection performance; for the ISCX-2014 dataset, the accuracy of the framework is less than 1% lower than the best effect.

Understanding what happens in surveillance videos is critical for human–machine interaction in the Internet of Things (IoT). Among key tasks, video moment localization (VML) aims to locate the moment within untrimmed videos that semantically corresponds to a given query. Current studies predominantly emphasize visual components of videos, often neglecting the rich semantic content in the accompanying audio. Moreover, these approaches typically employ fixed fusion strategies during both training and inference phases, leading to limited flexibility and adaptability due to rigid architectures. To address these challenges, we propose a novel Progressive Dynamic Interaction Network with Audio Supplement (PDIN). Specifically, this framework employs a graph fusion approach to integrate audio and visual information, while addressing potential discordance between the two modalities. The incorporation of audio data significantly enhances the utilization of visual features, particularly in scenarios where visual information may be obscured or incomplete. Additionally, a progressive dynamic interaction strategy is also designed, where each layer builds upon the outcomes of the previous layer. This approach systematically addresses semantic differences between modalities, thereby facilitating a more integrated and effective fusion process. The underlying principle of this strategy is transferable to other multimodal tasks. Extensive experiments on three benchmark datasets (ActivityNet Captions, Charades-STA and QVHighlights) demonstrate that PDIN consistently outperforms existing methods.

A botnet is a group of hijacked devices that conduct various cyberattacks, which is one of the most dangerous threats on the internet. Individuals or organizations can effectively detect botnets by analyzing abnormal behaviors in network traffic. Existing works focus on extracting the deterministic behavioral features, which highly rely on statistical features and existing botnet interaction structures, resulting in unsatisfactory detection accuracy, especially for unknown botnet traffic. The botnet detection method based on the original traffic bytes has more advantages in this regard, especially the use of mining payload information in the traffic to enhance the identification of abnormal botnet behavior is the focus of this study. In this paper, we propose a dual-mode botnet detection scheme, which takes the original traffic bytes as the object, one is to encode the implicit semantic relationship between the traffic bytes through a multi-layer Transformer encoder, and the other is the network traffic Image representation, the spatial relationship of traffic bytes is captured by a deep neural network, and then botnet detection is achieved by maximizing the mutual information between the two. We conduct comprehensive experiments with both known botnets and unknown botnets to evaluate our scheme. Experimental results show that for known botnets, our approach achieves 99.84% and 91.92% detection accuracy with CTU-13 and ISCX-2014 datasets, respectively, which is 3.04% and 2.54% more accurate compared with the state-of-art (DL). For unknown datasets, our scheme is 10.19% more accurate than the existing traffic representation.

Abstract A novel video summarisation method called the Hyperbolic Graph Convolutional Network (HVSN) is proposed, which addresses the challenges of summarising edited videos and capturing the semantic consistency of video shots at different time points. Unlike existing methods that use linear video sequences as input, HVSN leverages Hyperbolic Graph Convolutional Networks (HGCNs) and an adaptive graph convolutional adjacency matrix network to learn and aggregate features from video shots. Moreover, a feature fusion mechanism based on the attention mechanism is employed to facilitate cross‐module feature fusion. To evaluate the performance of the proposed method, experiments are conducted on two benchmark datasets, TVSum and SumMe. The results demonstrate that HVSN achieves state‐of‐the‐art performance, with F1‐scores of 62.04% and 50.26% on TVSum and SumMe, respectively. The use of HGCNs enables the model to better capture the complex spatial structures of video shots, and thus contributes to the improved performance of video summarisation.

Aurora is the typical ionosphere track generated by the interaction of solar wind and magnetosphere, and its detection is significant to study of space weather activity. Space-borne ultraviolet detectors, especially far ultraviolet band image detecting device, provide abundant detecting data. Based on the special morphology of ultraviolet aurora images, the combination of texture and shape features is utilized to extract the features of ultraviolet aurora images, and then a support vector machine (SVM) is employed to classify the auroras. The experiment based on ultraviolet aurora image data obtained by the Polar satellite shows the feasibility and effectiveness of our feature representation method.