A Scalable and Generalised Deep Learning Framework for Anomaly Detection in Surveillance Videos

International Journal of Intelligent Systems, Год журнала: 2025, Номер 2025(1)

Опубликована: Янв. 1, 2025

Anomaly detection in videos is challenging due to the complexity, noise, and diverse nature of activities such as violence, shoplifting, vandalism. While deep learning (DL) has shown excellent performance this area, existing approaches have struggled apply DL models across different anomaly tasks without extensive retraining. This repeated retraining time‐consuming, computationally intensive, unfair. To address limitation, a new framework introduced study, consisting three key components: transfer enhance feature generalization, model fusion improve representation, multitask classification generalize classifier multiple training from scratch when task introduced. The framework’s main advantage its ability requiring for each task. Empirical evaluations demonstrate effectiveness, achieving an accuracy 97.99% on RLVS (violence detection), 83.59% UCF dataset (shoplifting 88.37% both datasets using single Additionally, tested unseen dataset, achieved 87.25% 79.39% violence shoplifting datasets, respectively. study also utilises two explainability tools identify potential biases, ensuring robustness fairness. research represents first successful resolution generalization issue detection, marking significant advancement field.

Язык: Английский

---

A Comparative Study on Recent Automatic Data Fusion Methods

Computers, Год журнала: 2023, Номер 13(1), С. 13 - 13

Опубликована: Дек. 30, 2023

Automatic data fusion is an important field of machine learning that has been increasingly studied. The objective to improve the classification performance from several individual classifiers in terms accuracy and stability results. This paper presents a comparative study on recent methods. step can be applied at early and/or late stages procedure. Early consists combining features different sources or domains form observation vector before training classifiers. On contrary, results after testing stage. Late two setups, combination posterior probabilities (scores), which called soft fusion, decisions, hard fusion. A theoretical analysis conditions for applying three kinds (early, late, hard) introduced. Thus, we propose with schemes including weaknesses strengths state-of-the-art methods studied following perspectives: sensors, features, scores, decisions.

Язык: Английский

---

RepVGG-SimAM: An Efficient Bad Image Classification Method Based on RepVGG with Simple Parameter-Free Attention Module

Applied Sciences, Год журнала: 2023, Номер 13(21), С. 11925 - 11925

Опубликована: Окт. 31, 2023

With the rapid development of Internet technology, number global users is rapidly increasing, and scale also expanding. The huge system has accelerated spread bad information, including images. Bad images reflect vulgar culture Internet. They will not only pollute environment impact core society but endanger physical mental health young people. In addition, some criminals use to induce download software containing computer viruses, which greatly security cyberspace. Cyberspace governance faces enormous challenges. Most existing methods for classifying face problems such as low classification accuracy long inference times, these limitations are conducive effectively curbing reducing their harm. To address this issue, paper proposes a method (RepVGG-SimAM) based on RepVGG simple parameter-free attention mechanism (SimAM). This uses backbone network embeds SimAM in so that neural can obtain more effective information suppress useless information. We used pornographic publicly disclosed by data scientist Alexander Kim violent collected from internet construct dataset our experiment. experimental results prove proposed reach 94.5% images, false positive rate 4.3%, speed doubled compared with ResNet101 network. Our identify provide efficient powerful support cyberspace governance.

Язык: Английский

---

A Hybrid-Transformer-Based Cyber-Attack Detection in IoT Networks

International Journal of Interactive Mobile Technologies (iJIM), Год журнала: 2024, Номер 18(14), С. 90 - 102

Опубликована: Авг. 2, 2024

The concept of the Internet Things (IoT) is significant in today’s world and opens up new opportunities for several organizations. IoT solutions are proliferating fields such as self-driving cars, smart homes, transportation, healthcare, services constantly being created. Over previous decade, society has seen a expansion connectivity. In reality, connectivity will expand variety domains over next few years. Various problems must be overcome to permit effective secure operations. However, growing connections increase potential cyber-attacks since attackers can exploit broad network linked devices. Artificial intelligence (AI) detects prevents cyber assaults by developing adjusting threats weaknesses. this study, we offer novel cyber-detection model networks based on convolutional neural (CNN) transformers. study aims enhance system’s ability identify detect cyberattacks, sophisticated assaults, its performance. experimental findings, using cybersecurity CICIoT2023 dataset, show that CNN-Transformer hazards with an overall accuracy 99.49%. identifying hazardous activity, MLP 99.39%, while XGBoost-pipeline 99.40%.

Язык: Английский

---

Enhancing Smart City Safety and Utilizing AI Expert Systems for Violence Detection

Future Internet, Год журнала: 2024, Номер 16(2), С. 50 - 50

Опубликована: Янв. 31, 2024

Violent attacks have been one of the hot issues in recent years. In presence closed-circuit televisions (CCTVs) smart cities, there is an emerging challenge apprehending criminals, leading to a need for innovative solutions. this paper, propose model aimed at enhancing real-time emergency response capabilities and swiftly identifying criminals. This initiative aims foster safer environment better manage criminal activity within cities. The proposed architecture combines image-to-image stable diffusion with violence detection pose estimation approaches. generates synthetic data while object approach uses YOLO v7 identify violent objects like baseball bats, knives, pistols, complemented by MediaPipe action detection. Further, long short-term memory (LSTM) network classifies involving objects. Subsequently, ensemble consisting edge device entire deployed onto testing using dash camera. Thus, study can handle send alerts emergencies. As result, our achieves mean average precision (MAP) 89.5% attack detection, LSTM classifier accuracy 88.33% classification. results highlight model’s enhanced capability accurately detect objects, particularly effectively through implemented artificial intelligence system.

Язык: Английский

---

Automate facial paralysis detection using vgg architectures

International Journal of Current Innovations in Advanced Research, Год журнала: 2024, Номер unknown, С. 1 - 8

Опубликована: Фев. 9, 2024

Facial Paralysis (FP) is a debilitating condition that affects individuals worldwide by impairing their ability to control facial muscles and resulting in significant physical emotional challenges. Precise prompt identification of FP crucial for appropriate medical intervention treatment. With the advancements deep learning techniques, specifically Convolutional Neural Networks (CNNs), there has been growing interest utilising these models automated detection. This paper investigates effectiveness CNN architectures identify patients with paralysis. The proposed method leveraged depth simplicity Visual Geometry Group (VGG) capture intricate relationships within images accurately classify on YouTube Palsy (YFP) dataset. dataset consists 2000 categorised into non-injured individuals. Data augmentation techniques were used improve robustness generalisation approach proposed. model features extraction module VGG network classification Softmax classifier. performance evaluation metrics include accuracy, recall, precision F1-score. Experimental results demonstrate VGG16 scored an accuracy 88.47% recall 83.55%, 92.15% F1-score 87.64%. VGG19 attained level 81.95%, 72.44%, 88.58% 79.70%. outperformed terms precision, indicate are effective identifying

Язык: Английский

---

Deep Reinforcement Learning Approach for Cyberattack Detection

International Journal of Online and Biomedical Engineering (iJOE), Год журнала: 2024, Номер 20(05), С. 15 - 30

Опубликована: Март 15, 2024

Recently, there has been a growing concern regarding the detrimental effects of cyberattacks on both infrastructure and users. Conventional safety measures, such as encryption, firewalls, intrusion detection, are inadequate to safeguard cyber systems against emerging evolving threats. To address this issue, researchers have turned reinforcement learning (RL) potential solution for complex decision-making problems in cybersecurity. However, application RL faces various obstacles, including lack suitable training data, dynamic attack scenarios, challenges modeling real-world complexities. This paper suggests applying deep (DRL), framework, simulate malicious enhance Our framework utilizes an agent-based model that is capable continuous adaptation within network security environment. The agent determines most optimal course action based network’s state corresponding rewards received its decisions. We present outcomes our experimentation with DRL specific model, double Q-network (DDQN), utilizing policy gradient (PG) three distinct datasets: NSL-KDD, CIC-IDS-2018, AWID. research demonstrates can effectively improve cyberattack detection through parameter adjustments.

Язык: Английский

---

A survey about deep learning and federated Learning in cyberse-curity

Periodicals of Engineering and Natural Sciences (PEN), Год журнала: 2024, Номер 12(1), С. 75 - 75

Опубликована: Фев. 10, 2024

Advances in Artificial Intelligence (AI) technology have led to the strengthening of traditional systems' cybersecurity capabilities a variety applications. However, these embedded machine learning models exposed systems new set vulnerabilities known as AI assaults. These are now attractive targets for cyberattacks, jeopardizing security and safety bigger that include them. As result, DL approaches critical transitioning network system protection from providing safe communication between intelligence security. Federated (FL) is kind based on heterogeneous datasets decentralized training. FL unique research topic currently its early phases. It has not yet gained wide acceptance community, owing mostly privacy considerations. In this research, we first shed light risks must be discovered, analyzed, recorded. favored scenarios where paramount is-sues. An extensive understanding risk factors allows an adopter implementer construct environment successfully while giving researchers clear perspective possible study domains. The survey paper intends analysis modern advances improve enhanced methods. proposes complete examination FL's issues assist bridging gap current level federated future which broad adoption achievable. We also propose range most recently used rating standards.

Язык: Английский

---

Deep Learning for Abnormal Human Behavior Detection in Surveillance Videos - a Survey

Опубликована: Янв. 1, 2024

Detecting abnormal human behaviors in surveillance videos is crucial for various domains such as security and public safety. However, the scarcity of labeled behavior data poses significant challenges developing effective detection systems. This paper presents a comprehensive survey deep learning techniques detecting video streams. We categorize existing approaches into four categories: reconstruction-based, generative-based, partially-supervised-based, fully-supervised-based methods. Each approach examined terms its underlying conceptual framework, strengths, drawbacks. Additionally, we provide an extensive comparison these using popular datasets frequently employed prior research, highlighting their performance across different scenarios. summarize advantages disadvantages each detection. also discuss open research issues identified through our survey, including enhancing robustness to environmental variations diverse realistic datasets, formulating strategies contextual detection, addressing gradient exploding issue, designing lightweight models real-world Finally, outline potential directions future development pave way more

Язык: Английский

---

Internet of Things: Architecture, Technologies, Applications, and Challenges

Alkadhim journal for computer science., Год журнала: 2024, Номер 2(1), С. 36 - 52

Опубликована: Март 14, 2024

A subset of cutting-edge information technology is the Internet things (IoT). IoT refers to a network physical objects with sensors attached that are linked via LAN and WAN networking techniques. It now commonly used sense environment gather data in variety settings, including smart cities, healthcare, intelligent transportation, homes, other structures. The architecture, core technology, significant applications were outlined this overview. sensing layer, transport application layer separated architecture. essential technologies embedded systems, connectivity, sensor, radio frequency identification (RFID) technology. implementation logistics still faces challenges despite potential advantages. utilization context topic many open studies, which also examined paper.

Язык: Английский

---