Defense Scenario Research Articles

Deep learning adversarial attacks and defenses in autonomous vehicles: a systematic literature review from a safety perspective

The integration of Deep Learning (DL) algorithms in Autonomous Vehicles (AVs) has revolutionized their precision in navigating various driving scenarios, ranging from anti-fatigue safe driving to intelligent route planning. Despite their proven effectiveness, concerns regarding the safety and reliability of DL algorithms in AVs have emerged, particularly in light of the escalating threat of adversarial attacks, as emphasized by recent research. These digital or physical attacks present formidable challenges to AV safety, relying extensively on collecting and interpreting environmental data through integrated sensors and DL. This paper addresses this pressing issue through a systematic survey that meticulously explores robust adversarial attacks and defenses, specifically focusing on DL in AVs from a safety perspective. Going beyond a review of existing research papers on adversarial attacks and defenses, the paper introduces a safety scenarios taxonomy matrix Inspired by SOTIF designed to augment the safety of DL in AVs. This matrix categorizes safety scenarios into four distinct areas and classifies attacks into those areas in three scenarios, along with two defense scenarios. Furthermore, the paper investigates the testing and evaluation measurements critical for assessing attacks in the context of DL for AVs. It further explores the dynamic landscape of datasets and simulation platforms. This contribution significantly enriches the ongoing discourse surrounding the assurance of safety and reliability in autonomous vehicles, especially in the face of continually evolving adversarial challenges.

Influence of Fatigue and Defensive Pressure on Three-Point Jump-Shot Kinematics in Basketball

This study examines the influence of fatigue and defensive pressure on the kinematic parameters of the three-point jump shot in basketball. Fourteen male collegiate basketball players (age: 21 ± 3 years old, body height: 186.35 ± 7.02 cm, body mass: 82.20 ± 10.99) participated in the study. Each participant performed three-point jump shots under four conditions: without defense, with defense, without defense after a fatigue protocol, and with defense after a fatigue protocol. Kinematic data were collected using the Xsens MVN inertial suit system and the OptoJump Next system. The analysis focused on various parameters including jump height, center of mass, release height, shoulder angle, and segment velocities. The repeated-measures ANOVA was used to observe the differences between each shot condition (fatigue, defense). Results indicated significant changes in the kinematic parameters due to both fatigue and defensive pressure. Fatigue notably changed shooting performance, affecting jump height and release mechanics. The defensive pressure altered shooting technique, leading to quicker ball release and higher release points. These findings highlight the importance of incorporating fatigue and defensive scenarios in training, suggesting that coaches develop more targeted training plans to improve performance under conditions of fatigue and defensive pressure.

Interface Design to Support Transfer of Tactical Control (ToTC) in Distributed Teams

To be effective, future combined joint operations will require distributed teams to share assets via the Transfer of Tactical Control (ToTC). The goal of this research was to develop and conduct a usability assessment of ToTC enabling interfaces within a UxV command and control simulation testbed. Three interfaces were developed: Enhanced Play Workbook, ToTC Request Tile, and Enhanced Vehicle Status Tile. These interfaces were evaluated by eleven participants with UxV (MQ-9 pilots/sensor operators) or battle management experience within a base defense scenario. Within the resource constrained scenario, participant completed just over half of the tasks and ToTC request rates were low overall. When UxV requests were performed by participants, their feedback suggested that the interfaces enhanced team situation awareness, but the request process could be further streamlined to support situations in which there might be a high rate of requests.

A joint lightly perturbation and synthesizing optimization attack framework for point cloud perturbation

Due to the inherent characteristic of deep learning models, prior researchers have focused on studying adversarial strategies aimed at uncovering their vulnerabilities. However, these strategies have been primarily been confined to the realm of image processing. When considering 3D vision structures, such as point clouds and meshes, the robustness of learning models also diminishes significantly in the presence of discernible outliers, and these strategies often fail to generalize across different DNN architectures. In this paper, a simple adversarial perturbation and optimization attack framework, Tangent Plane Perturbation and Synthesizing Optimization Attack (TPSOA) is presented. TPSOA provides a point cloud projection direction for point clouds, subtly inducing imperceptible perturbations on their tangent planes. Furthermore, acknowledging the geometry form of adversarial generation and its potential for 3D printing in physical scenarios, we reconstruct the adversarial mesh alongside the adversarial point cloud. A synthesized loss, incorporating both the Sobolev loss and the Chamfer loss, is utilized to optimize the distances between the adversarial generation and ground truth. Through a combination of visualization and quantitative analysis experiments, we validate the imperceptible nature of TPSOA, the necessity of the loss terms, and the magnitude of perturbations. These experiments that TPSOA enables targeted point cloud attacks that are imperceptible and of minor magnitude. In real-world adversarial attack and defense scenarios, as evidenced by both white-box and black-box experiments, the results show that TPSOA surpasses other state-of-the-art adversarial attack models, effectively manipulating a greater number of victim models while maintaining high efficiency, all without compromising its original structural integrity.

Experimental and Numerical Simulation of Ejecta Size and Velocity of Hypervelocity Impact Rubble-Pile Asteroid

Rubble-pile asteroids may be the type of near-Earth object most likely to threaten Earth in a future collision event. Small-scale impact experiments and numerical simulations for large-scale impacts were conducted to clarify the size ratio of the boulder/projectile diameter effects on ejecta size–velocity distribution. A series of small-scale impact cratering experiments were performed on porous gypsum–basalt targets at velocities of 2.3 to 5.5 km·s−1. Three successive ejection processes were observed by high-speed and ultra-high-speed cameras. The momentum transfer coefficient and cratering size were measured. A three-dimensional numerical model reflecting the random distribution of the interior boulders of the rubble-pile structure asteroid is established. The size ratio (length to diameter) of the boulder size inside the asteroid to the projectile diameter changed from 0.25 to 1.7. We conducted a smoothed particle hydrodynamics numerical simulation in the AUTODYN software to study the boulder size effect on the ejecta size–velocity distribution. Simulation results suggest that the microscopic porosity on regolith affects the propagation of shock waves and reduces the velocity of ejecta. Experiments and numerical simulation results suggest that both excavation flow and spalling ejection mechanism can eject boulders (0.12–0.72 m) out of the rubble-pile asteroid. These experiments and simulation results help us select the potential impact site in a planetary defense scenario and reduce deflection risk. are comprised primarily of boulders of a range of sizes.

An Attacker-Defender Game Model with Constrained Strategies.

Recently, research interest in the field of infrastructure attack and defense scenarios has increased. Numerous methods have been proposed for studying strategy interactions that combine complex network theory and game theory. However, the unavoidable effect of constrained strategies in complex situations has not been considered in previous studies. This study introduces a novel approach to analyzing these interactions by including the effects of constrained strategies, a factor often neglected in traditional analyses. First, we introduce the rule of constraints on strategies, which depends on the average distance between selected nodes. As the average distance increases, the probability of choosing the corresponding strategy decreases. Second, we establish an attacker-defender game model with constrained strategies based on the above rule and using information theory to evaluate the uncertainty of these strategies. Finally, we present a method for solving this problem and conduct experiments based on a target network. The results highlight the unique characteristics of the Nash equilibrium when setting constraints, as these constraints influence decision makers' Nash equilibria. When considering the constrained strategies, both the attacker and the defender tend to select strategies with lower average distances. The effect of the constraints on their strategies becomes less apparent as the number of attackable or defendable nodes increases. This research advances the field by introducing a novel framework for examining strategic interactions in infrastructure defense and attack scenarios. By incorporating strategy constraints, our work offers a new perspective on the critical area of infrastructure security.

Attack–defense strategy of UAV swarm based on DEP-SIQ in the active target defense scenario

Attack–defense strategy of UAV swarm based on DEP-SIQ in the active target defense scenario

Adversarial Machine Learning in Industry: A Systematic Literature Review

Adversarial Machine Learning (AML) discusses the act of attacking and defending Machine Learning (ML) Models, an essential building block of Artificial Intelligence (AI). ML is applied in many software-intensive products and services and introduces new opportunities and security challenges. AI and ML will gain even more attention from the industry in the future, but threats caused by already-discovered attacks specifically targeting ML models are either overseen, ignored, or mishandled. Current AML research investigates attack and defense scenarios for ML in different industrial settings with a varying degree of maturity with regard to academic rigor and practical relevance. However, to the best of our knowledge, a synthesis of the state of academic rigor and practical relevance is missing. This literature study reviews studies in the area of AML in the context of industry, measuring and analyzing each study’s rigor and relevance scores. Overall, all studies scored a high rigor score and a low relevance score, indicating that the studies are thoroughly designed and documented but miss the opportunity to include touch points relatable for practitioners.

Game-Based Learning in Modern Military Education: Digital Wargaming as a Tool for Tactical Training

Abstract This study examines the effectiveness of digital game-based scenarios and wargaming in developing tactical skills and decision-making abilities among military cadets. Utilizing the ARMA 3 game, three tactical scenarios—offensive, defensive, and stability and support operations—were tested. The study involved cadets from “Nicolae Bălcescu” Land Forces Academy (NBLFA) Sibiu, selected for their interest in gaming, familiarity with ARMA 3, and knowledge of military tactics. The results showed multiple advantages of using game-based scenarios in military training. The immersive environment provided by ARMA 3 enabled cadets to simulate realistic combat situations, enhancing their involvement and enthusiasm, leading to better knowledge absorption and skills development. Wargaming offered a safe learning environment, eliminating the risks of real exercises while allowing cadets to experience specific events and incidents. In the offensive mission, cadets faced challenges due to incomplete map analysis, highlighting the need to complement map analysis with real terrain reconnaissance. The defensive scenario underscored the importance of flexibility and adaptability of the combat forces’ disposition. In the stability and support scenario, cadets effectively managed civilian-military interactions, demonstrating the potential of digital games to develop broader competencies in multicultural communication and negotiation. The study concludes that digital game-based scenarios effectively prepare leaders for contemporary conflicts marked by conventional or hybrid threats. It recommends implementing a specific wargaming discipline in the military higher education curriculum and extending this approach to Land Forces units for continuous professional development.

Enhancing music rhythmic perception and performance with a VR game

This study analyzes the effect of using a virtual reality (VR) game as a complementary tool to improve users’ rhythmic performance and perception in a remote and self-learning environment. In recent years, remote learning has gained importance due to various everyday situations; however, the effects of using VR in such situations for individual and self-learning have yet to be evaluated. In music education, learning processes are usually heavily dependent on face-to-face communication with a teacher and are based on a formal or informal curriculum. The aim of this study is to investigate the potential of gamified VR learning and its influence on users’ rhythmic sensory and perceptual abilities. We developed a drum-playing game based on a tower defense scenario designed to improve four aspects of rhythmic perceptual skills in elementary school children with various levels of music learning experience. In this study, 14 elementary school children received Meta Quest 2 headsets for individual use in a 14-day individual training session. The results showed a significant increase in their rhythmical skills through an analysis of their rhythmic performance before and after the training sessions. In addition, the experience of playing the VR game and using the HMD setup was also assessed, highlighting some of the challenges of currently available affordable headsets for gamified learning scenarios.

Turing-like Experiment in a Cyber Defense Game

During the past decade, researchers of behavioral cyber security have created cognitive agents that are able to learn and make decisions in dynamic environments in ways that assimilate human decision processes. However, many of these efforts have been limited to simple detection tasks and represent basic cognitive functions rather than a whole set of cognitive capabilities required in dynamic cyber defense scenarios. Our current work aims at advancing the development of cognitive agents that learn and make defense-dynamic decisions during cyber attacks by intelligent attack agents. We also aim to evaluate the capability of these cognitive models in ``Turing-like'' experiments, comparing the decisions and performance of these agents against human cyber defenders. In this paper, we present an initial demonstration of a cognitive model of the defender that relies on a cognitive theory of dynamic decision-making, Instance-Based Learning Theory (IBLT); we also demonstrate the execution of the same defense task by human defenders. We rely on OpenAI Gym and CybORG and adapt an existing CAGE scenario to generate a simulation experiment using an IBL defender. We also offer a new Interactive Defense Game (IDG), where \textit{human} defenders can perform the same CAGE scenario simulated with the IBL model. Our results suggest that the IBL model makes decisions against two intelligent attack agents that are similar to those observed in a subsequent human experiment. We conclude with a description of the cognitive foundations required to build autonomous intelligent cyber defense agents that can collaborate with humans in autonomous cyber defense teams.

Disulfide-adducts with cysteine residues in human serum albumin prove exposure to malodorous mercaptans in vitro

Malodorants are mixtures containing mercaptans, which trigger the flight instinct upon exposure and might thus be deployed in military and civilian defense scenarios. Exposure to mercaptans might lead to unconsciousness, thus representing a possible threat for health. Therefore, we developed and validated a bioanalytical procedure for the simultaneous detection and identification of corresponding biomarkers for the verification of exposure to mercaptans. Disulfide-adducts of ethyl mercaptan (SEt), n-butyl mercaptan (SnBu), tert-butyl mercaptan (StBu) and iso-amyl mercaptan (SiAm) with cysteine (Cys) residues in human serum albumin (HSA) were formed by in vitro incubation of human plasma. After pronase-catalyzed proteolysis, reaction products were identified as adducts of the single amino acid Cys and the dipeptide cysteine-proline (Cys34Pro) detected by a sensitive μLC-ESI MS/MS method working in the scheduled multiple reaction monitoring (sMRM) mode. Dose-response studies showed linearity for the yield of Cys34Pro-adducts in the range from 6 nM to 300 μM of mercaptans in plasma and limits of identification (LOI) were in the range from 60 nM to 6 μM. Cys34-adducts showed stability for at least 6 days in plasma (37 °C). The presented disulfide-biomarkers expand the spectrum for bioanalytical verification procedures and might be helpful to prove exposure to malodorants.

TARGET RECOGINITION AND DETECTION USING DEEP LEARING TECHNIQUE

Object detection is a fundamental computer vision function for identifying and locating things in images or videos. Unlike image classification, object detection not only classifies the items in an image but also locates them inside the picture by constructing a bounding box around each one. Defence objects include a wide range of key assets needed for military and security activities. Tanks, armoured personnel carriers (APCs), military trucks, planes, helicopters, naval boats, and unmanned ground vehicles (UGVs) are among the types of vehicles included. Detection of these vehicles is critical for tracking troop movements, analysing battlefield dynamics, and maintaining operational preparedness. The modern defence scenario demands improved surveillance and security measures to successfully counter growing threats. This work suggests using the You Only Look Once (YOLO) method for real-time identification of numerous defence items. The project's goal is to improve security by properly detecting a variety of items, including cars, individuals, equipment, and structures, with a single deep learning model. Methodologically, a broad dataset containing photos and videos of defence items in various locations is collected and preprocessed. The YOLO algorithm is then trained on this dataset, with parameters optimised for high accuracy and recall rates. Performance evaluation include thorough testing on various datasets to determine accuracy, speed, and resilience. KEYWORDS: Target recognition, Detection, YOLO algorithm, Military applications, Flying objects, Vehicles, Sensor integration, Real-time processing, Threat detection, Situational awareness.

Consequences of asteroid characterization on the state of knowledge about inferred physical parameters and impact risk

Planetary defense scenarios require an assessment of the possible effects of an impact and of the efficacy of different mitigation options in order to provide decision makers with actionable information. Essential inputs to these assessments are the orbital parameters, the likelihood of impact, the possible locations of impact, and the physical properties of the potential impactor. There is a robust framework in place to determine the orbital parameters, the likelihood of impact, and the possible locations of impact (e.g. Milani et al. [1], Sciarra et al. [2], Losacco et al. [3]). While there is a robust suite of potential methods to determine properties of a potential impactor, most of these methods depend critically on the details of the objects observability (e.g. brightness, distance from Earth, …). As a result, in practice it is rare that all the physical properties of interest are measurable. In order to fill this gap, we have developed a property inference network that can be used to generate a set of potential impactors whose properties in aggregate are consistent with the distributions from the available measurements, the distributions derived from population-wide statistics, and the combination of physical properties for each individual impactor is physically plausible. We will describe this Asteroid Property Inference Network (APIN), apply it to the PDC23 scenario, and demonstrate how even limited asteroid characterization can be leveraged to improve impact risk assessment.

Motif-Backdoor: Rethinking the Backdoor Attack on Graph Neural Networks via Motifs

Graph neural network (GNN) with a powerful representation capability has been widely applied to various areas. Recent works have exposed that GNN is vulnerable to the backdoor attack, i.e., models trained with maliciously crafted training samples are easily fooled by patched samples. Most of the proposed studies launch the backdoor attack using a trigger that is either the randomly generated subgraph e.g., erdős-rényi backdoor (ER-B) for less computational burden or the gradient-based generative subgraph e.g., graph trojaning attack (GTA) to enable a more effective attack. However, the interpretation of how is the trigger structure and the effect of the backdoor attack related has been overlooked in the current literature. Motifs, recurrent and statistically significant subgraphs in graphs, contain rich structure information. In this article, we are rethinking the trigger from the perspective of motifs and propose a motif-based backdoor attack, denoted as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Motif-Backdoor</i> . It contributes from three aspects. First, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Interpretation</i> , it provides an in-depth explanation for backdoor effectiveness by the validity of the trigger structure from motifs, leading to some novel insights, e.g., using subgraphs that appear less frequently in the graph as the trigger can achieve better attack performance. Second, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Effectiveness</i> , Motif-Backdoor reaches the state-of-the-art (SOTA) attack performance in both black-box and defensive scenarios. Third, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Efficiency</i> , based on the graph motif distribution, Motif-Backdoor can quickly obtain an effective trigger structure without target model feedback or subgraph model generation. Extensive experimental results show that Motif-Backdoor realizes the SOTA performance on three popular models and four public datasets compared with five baselines, e.g., Motif-Backdoor improves the attack success rate (ASR) by 14.73% compared with baselines on average. In addition, under a possible defense, Motif-Backdoor still implements satisfying performance, highlighting the requirement of defenses against backdoor attacks on GNNs. The datasets and code are available at https://github.com/Seaocn/Motif-Backdoor.

Extracting Features from Foul Actions of Basketball Players in Real Time Using Machine Vision

As the global economic landscape rapidly advances and people's quality of life continues to improve, there is a noticeable increase in the demand for various sports, including basketball. Athletes of diverse skill levels find physical, social, and emotional satisfaction in engaging in basketball activities. In contemporary times, the requisites for success in basketball, such as height, speed, accuracy, distance, and strength, are escalating. The frequency of body contact has also intensified, leading to heightened intensity in ball shooting among players and an inevitability of fouls in offensive and defensive interactions. This dynamic poses challenges to the seamless progression of the sport, and the existence of various foul rules adds an element of unpredictability and uncontrollability to the game of basketball. The data reveals interesting insights. In terms of dribble defense, the Chinese team committed 16 fouls, whereas their opponents committed 33 fouls. However, in the context of shooting defense, the Chinese team had a higher foul count compared to their opponents. Conversely, when it came to defending the ball, the Chinese team committed fewer fouls than their opponents. In the category of preventing the ball, the Chinese team had 16 fouls compared to the opponent team's 12 fouls. Overall, the Chinese team demonstrated fewer fouls than their opponents in the realm of defense. The data underscores a potential highlight the specific factors influencing foul occurrences in different defensive scenarios. The study introduces a machine vision approach, which outperformed existing methods in classifying actions as fouls or not. This innovative application showcases the effectiveness of advanced technology in addressing the challenges associated with fouls in basketball, offering a promising avenue for further research and practical implementation. The comprehensive experimental results suggest that the machine vision approach outperformed existing methods in classifying actions as fouls or not, showcasing its effectiveness in addressing the challenge of fouls in basketball.

Optimization of high-speed fixed-wing UAV penetration strategy based on deep reinforcement learning

The penetration decision of unmanned aerial vehicles (UAVs) is one of the key components for UAVs to detect or strike crucial targets in adversarial environments. Presently, most methods lack research on penetration decisions under unknown interception information. The executing UAV for decision-making is generally a simplified three-degree-of-freedom model. The decision format is often simplified to path point optimization, lacking control over the stability of complex model attitudes and exhibiting relatively poor robustness. This paper proposes a penetration relative motion theory and aircraft control method based on the six-degrees-of-freedom UAV model to reduce response errors between algorithm control decisions and actual flight control. Using a Markov decision process, a UAV penetration decision control method based on an attitude-overload control loop is designed. This method combines reinforcement learning techniques to enable autonomous penetration decision-making and control for UAVs facing active defense scenarios, enhancing the autonomy, accuracy, and generalization of UAV online decision-making. The paper concludes by implementing autonomous decision-making and control for UAV penetration through the integration of interceptor trajectory prediction with proximal policy optimization. This method reduces the dependency of penetration strategies on interceptor information, ensuring both the generalization ability of penetration strategy models under uncertain information and the consistency of decision stability. Through simulations and experiments, we verified that the penetration strategy based on the pre-sampling PPO method improves the penetration effect by 15.44 % on average and the hit rate by 18.48 %. Moreover, the probability of penetration in different scenarios exceeds 70 %, and the penetration effect is improved by more than 10 % in the case of multiple interceptors. This paper also analyzes the impact of interceptor spatial distribution and the quantity of interceptors on the effectiveness of UAV penetration decision-making.

“MALVA ALCEA”: ADVANCING UKRAINIAN NAVALCAPABILITIES WITH REACT NATIVE TECHNOLOGY

In the rapidly evolving domain of military operations, the advent of digital technologies has ushered in unprecedented capabilities and efficiencies. “The Digital Admiral “Malva Alcea” exemplifies this transformation, showcasing a React Native-based mobile application designed to revolutionize missile targeting calculations within the Ukrainian Navy. This paper delves into the critical need for innovation in the face of traditional, manual calculation methods that are both time-consuming and prone to human error. It charts the development journey of the application, highlighting the selection of React Native for its cross-platform capabilities and ease of integration with naval data systems. The paper provides a comprehensive overview of the application’s architecture, including its intuitive user interface, robust storage model, and the integration of a sophisticated mathematical model for missile targeting. Through user feedback and operational deployment, the application has demonstrated a significant reduction in calculation times and an improvement in accuracy, thus enhancing operational efficiency and decision-making in naval defense scenarios. Furthermore, the paper addresses the challenges encountered during development, such as ensuring data security and maintaining system adaptability for future expansions. It concludes by outlining future directions for the application, including potential integrations with larger situational awareness platforms and expansions to other operating systems, underscoring the pivotal role of digital innovations in advancing military capabilities and strategic operations. Keywords: React Native, Naval Defense Operations, Missile Targeting Calculations, Mobile Application Development, User Experience (UX), Data Security and Encryption, Redux State Management, Encrypted Storage, Operational Efficiency, Military Technology Innovation

Characterization of the DART Impact Ejecta Plume on Dimorphos from LICIACube Observations

We modeled the geometry and the three-dimensional orientation of the ejecta cone triggered by the impact of the DART spacecraft on the asteroid Dimorphos. We used eight LUKE images of the impact acquired by the CubeSat LICIACube that flew by the Didymos system shortly after the impact. These images, which show the ejecta cone in both face-on and side-on profiles, enabled us to reconstruct the ejecta cone in inertial space. We started our model as a simple cone with a circular base and developed it to a rotated cone with an elliptical base that best fit the data. The cone axis points to R.A., decl. (in J2000): , +. The cone is characterized by two perpendicular half-angles of and a rotation of ω = 12° around its axis. The apex of the cone is located near the center of Dimorphos within 15 m. The intersection of the cone and the surface of Dimorphos (surface enclosed by the cone) would correspond to a crater with a maximum radius of about 65 m. The characterization of the cone axis is directly related to the computation of the momentum enhancement factor (β) of the impact, and it hence proves the crucial need of studying impacts in the context of planetary defence scenarios. The results of this work could potentially be used to constrain whether the impact took place in a strength-dominated or a gravity-dominated regime. This work shows the important scientific return of the LICIACube CubeSat in the context of planetary defence.

Effective defense strategies in network security using improved double dueling deep Q-network

Network security is a critical discipline in the contemporary digital world, encompassing diverse technologies and strategies aimed at safeguarding computer systems, networks, and data resources against malicious activities. The attackers and defenders are vital components in the context of network security and defense. Attackers employ various means to steal sensitive information, compromise system functionality, and potentially lead to substantial economic and societal damages. To address these challenges, various network attack and defense scenarios were constructed within the CybORG framework in this paper. In various scenarios, attacks were carried out by different attackers. This was done to investigate the diverse strategies employed by defenders in response to network intrusions across different scenarios. Additionally, real-time assessments of the effectiveness of defensive measures were conducted. To assess the efficacy of defense strategies, we propose DDQN-Dueling-Noisy-Experience Replay (DDQN-DNER), a deep reinforcement learning algorithm that trains the defense agent to take appropriate actions to protect the network as it transitions into various states of being under attack. Built upon the Deep Q-Network (DQN) algorithm, the DDQN-DNER method incorporates noise networks, additional experience replay, and distinguishes outputs into value functions and advantage functions, enabling the proactive updating of Q-network parameters based on optimal actions. Simultaneously, Gaussian noise is incorporated into the actions undertaken by the agents. Research findings indicate that as network complexity increases, it becomes more challenging for agents to formulate effective strategies, while lower network security enhances agent capability in strategic decision-making. Compared to the DDQN algorithm, the DDQN-DNER algorithm accelerates the convergence of the model. In all scenarios, this algorithm consistently achieves the highest scores, indicating that the defensive strategies and measures generated by the blue agent are highly effective. The blue agent can promptly detect potential threats and attacks and take appropriate actions to address and mitigate these attacks, thereby ensuring network security.