ÖZEL GÜVENLİK GÖZETİMİNDE DAVRANIŞ TANIMA İÇİN CNN MİMARİLERİNİN KARŞILAŞTIRMALI DEĞERLENDİRMESİ

Öz

Derin öğrenmenin özel güvenliğin geliştirilmesindeki rolü, akıllı video gözetim sistemlerinin entegrasyonu yoluyla önemli ölçüde güçlendirilmiştir. Bu çalışmada, özel güvenlikle ilgili çok sınıflı bir davranış tanıma görevi bağlamında üç evrişimsel sinir ağı (CNN) mimarisi — EfficientNet-B0, ResNet50 ve MobileNetV2 — değerlendirilmiştir. Kötüye kullanım, hırsızlık ve mağaza hırsızlığı gibi davranışlar, dokuz yasa dışı davranış ve bir normal aktiviteyi temsil eden on farklı sınıftan oluşan bir veri kümesi kullanılarak gözetim video verilerinden tespit edilmiştir. CNN modellerini bu sınıflandırma görevine uyarlamak için transfer öğrenme teknikleri uygulanmıştır. Model performansları kesinlik, duyarlılık, F1-skoru, karmaşıklık matrisleri ve çok sınıflı ROC eğrilerine dayalı olarak değerlendirilmiştir. EfficientNet-B0 modelinin, daha derin ResNet50 (%96,04 doğruluk) ve hafif MobileNetV2 (%96,54 doğruluk) modellerini aşarak test kümesinde en yüksek doğruluk (%98,11) ve F1-skoru (~0,98) elde ettiği gözlemlenmiştir. EfficientNet-B0 tarafından tüm sınıflar genelinde ROC eğrilerinin altında neredeyse mükemmel alanlar elde edilmiştir; bu durum, güçlü tespit yeteneklerini göstermektedir. ResNet50 ve MobileNetV2 de tatmin edici performans sergilemiş olsa da ateşli silah kullanımı veya soygun içeren belirli davranış kategorileri için daha düşük duyarlılık değerleri kaydedilmiştir. Bu bulguların özel güvenlik açısından sonuçları tartışılmış; özellikle EfficientNet-B0’ın, gözetim bağlamlarında gerçek zamanlı ve yüksek doğruluklu otomatik davranış tespiti için uygulanabilir bir çözüm olduğu öne sürülmüştür. Sonuç olarak, modern CNN mimarilerinin — özellikle EfficientNet’in — özel güvenlikte gözetim analitiğini geliştirmek için etkili araçlar sağladığı ve proaktif tehdit tanımlama ile suç önleme sistemlerinde kullanılmak üzere umut verici bir potansiyele sahip olduğu sonucuna varılmıştır.

Anahtar Kelimeler

• Özel Güvenlik
• Video Gözetimi
• Davranış Tanıma
• EfficientNet
• ResNet
• MobileNet

COMPARATIVE EVALUATION OF CNN ARCHITECTURES FOR BEHAVIOR RECOGNITION IN PRIVATE SECURITY SURVEILLANCE

Öz

The role of deep learning in enhancing private security has been significantly reinforced through the integration of intelligent video surveillance systems. In this study, three convolutional neural network (CNN) architectures — EfficientNet-B0, ResNet50, and MobileNetV2 — have been evaluated within the context of a multi-class behavior recognition task pertinent to private security. Behaviors such as abuse, burglary, and shoplifting were detected from surveillance video data by utilizing a dataset comprising ten distinct classes (nine illicit behaviors and one representing normal activity). Transfer learning techniques have been applied to adapt the CNN models for this classification task. Model performances have been assessed based on precision, recall, F1-score, confusion matrices, and multi-class ROC curves. The EfficientNet-B0 model was observed to yield the highest accuracy (98.11%) and F1-score (~0.98) on the test set, surpassing both the deeper ResNet50 (96.04% accuracy) and the lightweight MobileNetV2 (96.54% accuracy). Near-perfect area under the ROC curves were achieved by EfficientNet-B0 across all classes, indicating robust detection capabilities. Although ResNet50 and MobileNetV2 also demonstrated satisfactory performance, lower recall values were noted for specific behavior categories, such as those involving firearm use or robbery. The implications of these findings for private security have been discussed; in particular, the deployment of EfficientNet-B0 is suggested as a viable solution for real-time and highly accurate automated behavior detection in surveillance contexts. It is concluded that modern CNN architectures, especially EfficientNet, provide effective tools for advancing surveillance analytics within private security, with promising potential for use in proactive threat identification and crime prevention systems.

Anahtar Kelimeler

• Private Security
• Video Surveillance
• Behavior Recognition
• EfficientNet
• ResNet
• MobileNet

Kaynakça

1. AbuAlkebash, H., Saleh, R. A. A., & Ertunç, H. M. (2025). Automated explainable deep learning framework for multiclass skin cancer detection and classification using hybrid YOLOv8 and vision transformer (ViT). Biomedical Signal Processing and Control, 108, 107934. https://doi.org/10.1016/j.bspc.2025.107934
2. Ahmed, I., & Naib, B. B. (2023, 29–30 April 2023). Object Motion Tracking and Detection in Surveillance Videos using Resnet Architecture. 2023 International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE),
3. Akshith, N., Adithya, S. K., Abhiram, M., Vikas, M., Achuth, D., & Murthy, G. N. (2022, 22–23 Dec. 2022). Action Recognition with Neural Networks. 2022 Sardar Patel International Conference on Industry 4.0 - Nascent Technologies and Sustainability for 'Make in India' Initiative,
4. Al-Gaashani, M. S. A. M., Xu, W., & Obsie, E. Y. (2025). MobileNetV2-based deep learning architecture with progressive transfer learning for accurate monkeypox detection. Applied Soft Computing, 169, 112553. https://doi.org/10.1016/j.asoc.2024.112553
5. Alamuru, S., & Jain, S. (2022, 29–30 July 2022). A Quick Review and Performance Analysis of Custom and Transfer Learning CNN Architectures for Event Detection in Videos. 2022 IEEE International Conference on Data Science and Information System (ICDSIS),
6. Altal, O. F., Sindiani, A. M., Amin, M., Abu Mhanna, H. Y., Hamad, R., Gharaibeh, H., Akhdar, H. F., Alhatamleh, S., Almahmoud, R. E., Abu-azzam, O. H., Balaw, M., Hamoud, B. H., Maashey, F., & Alghulayqah, L. (2025). Hybrid attention-enhanced MobileNetV2 with particle swarm optimization for endometrial cancer classification in CT images. Informatics in Medicine Unlocked, 57, 101662. https://doi.org/10.1016/j.imu.2025.101662
7. Amin, S. U., Abbas, M. S., Kim, B., Jung, Y., & Seo, S. (2024). Enhanced Anomaly Detection in Pandemic Surveillance Videos: An Attention Approach With EfficientNet-B0 and CBAM Integration. IEEE Access, 12, 162697–162712. https://doi.org/10.1109/ACCESS.2024.3488797
8. Aminanto, M. E., Purbomukti, I. R., Chandra, H., & Kim, K. (2022). Two-Dimensional Projection-Based Wireless Intrusion Classification Using Lightweight EfficientNet. Computers, Materials and Continua, 72(3), 5301–5314. https://doi.org/10.32604/cmc.2022.026749

9. Ansari, M. A., & Singh, D. K. (2022). An Expert Eye for Identifying Shoplifters in Mega Stores. In A. Khanna, D. Gupta, S. Bhattacharyya, A. E. Hassanien, S. Anand, & A. Jaiswal, International Conference on Innovative Computing and Communications Singapore.
10. Argho, A. G., Maswood, M. M. S., Mahmood, M. I., & Mondol, N. (2024). EfficientCovNet: A CNN-based approach to detect various pulmonary diseases including COVID-19 using modified EfficientNet. Intelligent Systems with Applications, 21, 200315. https://doi.org/10.1016/j.iswa.2023.200315
11. Bhimavarapu, J. P., Ramaraju, S., Nagajyothi, D., & Rao, I. V. (2023). Convolutional neural network based object detection system for video surveillance application. Concurrency and Computation: Practice and Experience, 35(3), e7461. https://doi.org/10.1002/cpe.7461
12. Bohlol, P., Hosseinpour, S., & Soltani Firouz, M. (2025). Improved food recognition using a refined ResNet50 architecture with improved fully connected layers. Current Research in Food Science, 10, 101005. https://doi.org/10.1016/j.crfs.2025.101005
13. Canayaz, M. (2021). C+EffxNet: A novel hybrid approach for COVID-19 diagnosis on CT images based on CBAM and EfficientNet. Chaos, Solitons & Fractals, 151, 111310. https://doi.org/10.1016/j.chaos.2021.111310
14. Duong, H. T., Le, V. T., & Hoang, V. T. (2023). Deep Learning-Based Anomaly Detection in Video Surveillance: A Survey. Sensors (Basel), 23(11). https://doi.org/10.3390/s23115024
15. Elmetwally, A., Eldeeb, R., & Elmougy, S. (2025). Deep learning based anomaly detection in real-time video. Multimedia Tools and Applications, 84(11), 9555–9571. https://doi.org/10.1007/s11042-024-19116-9
16. Himel, G. M. S., Islam, M. M., & Rahaman, M. (2024). Utilizing EfficientNet for sheep breed identification in low-resolution images. Systems and Soft Computing, 6, 200093. https://doi.org/10.1016/j.sasc.2024.200093
17. Jaggi, A., Aggarwal, A., & Gupta, A. (2023, 3–4 March 2023). Identifying Anti-Social Activities in Surveillance Monitoring Applications using Deep-CNN based Algorithms. 2023 6th International Conference on Information Systems and Computer Networks (ISCON),
18. Kaggle. (2025). Real Time Anomaly Detection in CCTV Surveillance. https://www.kaggle.com/datasets/webadvisor/real-time-anomaly-detection-in-cctv-surveillance
19. Kerachi, S., Koma, A. K., & Asharioun, H. (2023, 25–26 Jan. 2023). Real-time Steal Recognition on CCTV-Based Videos for Embedded Systems. 2023 28th International Computer Conference, Computer Society of Iran (CSICC),
20. Khanam, S., Sharif, M., Cheng, X., & Kadry, S. (2024). Suspicious action recognition in surveillance based on handcrafted and deep learning methods: A survey of the state of the art. Computers and Electrical Engineering, 120, 109811. https://doi.org/10.1016/j.compeleceng.2024.109811
21. Li, J., Zeng, H., Huang, C., Wu, L., Ma, J., Zhou, B., Ye, D., & Weng, H. (2023). Noninvasive Detection of Salt Stress in Cotton Seedlings by Combining Multicolor Fluorescence–Multispectral Reflectance Imaging with EfficientNet-OB2. Plant Phenomics, 5, 0125. https://doi.org/10.34133/plantphenomics.0125
22. Li, Z., Li, Y., Yan, C., Yan, P., Li, X., Yu, M., Wen, T., & Xie, B. (2024). Enhancing Tea Leaf Disease Identification with Lightweight MobileNetV2. Computers, Materials and Continua, 80(1), 679–694. https://doi.org/10.32604/cmc.2024.051526
23. Liao, C., Pu, C., Chen, T., Todo, Y., Furuichi, K., Yabe, T., & Qiu, D. (2025). Image analysis of diabetes pathology: classifying with precision via an upgraded resnet50 framework. Medicine in Novel Technology and Devices, 27, 100385. https://doi.org/10.1016/j.medntd.2025.100385
24. Lu, Y., Xu, Z., & Wang, J. (2022, 24–26 June 2022). Abnormal Behavior Recognition System based on Improved CRNN Model. 2022 IEEE International Conference on Artificial Intelligence and Computer Applications (ICAICA),
25. Mahjoubi, M. A., Lamrani, D., Saleh, S., Moutaouakil, W., Ouhmida, A., Hamida, S., Cherradi, B., & Raihani, A. (2025).
26. Optimizing ResNet50 performance using stochastic gradient descent on MRI images for Alzheimer's disease classification. Intelligence-Based Medicine, 11, 100219. https://doi.org/10.1016/j.ibmed.2025.100219
27. Mohanaprakash, T. A., Somu, C. S., Nirmalrani, V., Vyshnavi, K., Sasikumar, A. N., & Shanthi, P. (2024, 18–20 Sept. 2024). Detection of Abnormal Human Behavior using YOLO and CNN for Enhanced Surveillance. 2024 5th International Conference on Smart Electronics and Communication (ICOSEC),
28. Muthulakshmi, M., Venkatesan, K., Prasanalakshmi, B., Syarifah Bahiyah, R., & Divya, V. (2025). An intelligent ensemble EfficientNet prediction system for interpretations of cardiac magnetic resonance images in heart failure severity diagnosis. Intelligence-Based Medicine, 11, 100218. https://doi.org/10.1016/j.ibmed.2025.100218
29. Natha, S., Ahmed, F., Siraj, M., Lagari, M., Altamimi, M., & Chandio, A. A. (2025). Deep BiLSTM Attention Model for Spatial and Temporal Anomaly Detection in Video Surveillance. Sensors, 25(1), 251. https://doi.org/10.3390/s25010251
30. Ngoc, H. N., Xuan, N. N., Bui, T. H., Hung, D. H., Truong, S. Q. H., & Hoang, V. (2023, 5–8 Jan. 2023). An efficient approach for real-time abnormal human behavior recognition on surveillance cameras. 2023 IEEE 17th International Conference on Automatic Face and Gesture Recognition (FG),
31. P. Gangan, M., K, A., & V. L, L. (2022). Distinguishing natural and computer generated images using Multi-Colorspace fused EfficientNet. Journal of Information Security and Applications, 68, 103261. https://doi.org/10.1016/j.jisa.2022.103261
32. Pathirannahalage, I., Jayasooriya, V., Samarabandu, J., & Subasinghe, A. (2025). A comprehensive analysis of real-time video anomaly detection methods for human and vehicular movement. Multimedia Tools and Applications, 84(10), 7519–7564. https://doi.org/10.1007/s11042-024-19204-w
33. Peng, L., Zhang, J., Li, Y., & Du, G. (2024). A novel percussion-based approach for pipeline leakage detection with improved MobileNetV2. Engineering Applications of Artificial Intelligence, 133, 108537. https://doi.org/10.1016/j.engappai.2024.108537
34. Rath, A., Mishra, B. S. P., & Bagal, D. K. (2025). ResNet50-based Deep Learning model for accurate brain tumor detection in MRI scans. Next Research, 2(1), 100104. https://doi.org/10.1016/j.nexres.2024.100104
35. Rautaray, J., Ali, A. B. M., Kandpal, M., Mishra, P., Rashid, R. F., Alimova, F., Kallel, M., & Batool, N. (2025). Leveraging FastViT based knowledge distillation with EfficientNet-B0 for diabetic retinopathy severity classification. SLAS Technology, 33, 100325. https://doi.org/10.1016/j.slast.2025.100325
36. Razak, H., Tahir, N. M., Abd Almisreb, A., Zakaria, N., & Zamri, N. (2022). Detection of criminal behavior at the residential unit based on deep convolutional neural network. International Journal of Advanced Computer Science and Applications, 13(2).
37. Redondo, A., Ivaylova, K., Bachiller, M., Rincón, M., Cuadra, J. M., Tamimi, F., López-Cedrún, J. L., Diniz-Freitas, M., Lago-Méndez, L., Rubín-Roger, G., Torres, J., Bagán, L., Hernández, G., & López-Pintor, R. M. (2026). Multiclass classification of oral mucosal lesions by deep learning from clinical images without performing any restrictions. Biomedical Signal Processing and Control, 111, 108337. https://doi.org/10.1016/j.bspc.2025.108337
38. Rokhva, S., Teimourpour, B., & Soltani, A. H. (2024). Computer vision in the food industry: Accurate, real-time, and automatic food recognition with pretrained MobileNetV2. Food and Humanity, 3, 100378. https://doi.org/10.1016/j.foohum.2024.100378
39. Saket, O., Aicha, Anis B., & Fathallah, H. (2025). Deep learning applied for abnormal human behavior recognition in video surveillance systems: A systematic review. Applied Intelligence, 55(13), 904. https://doi.org/10.1007/s10489-025-06797-4
40. Silva, D., Manzo-Martínez, A., Gaxiola, F., Gonzalez-Gurrola, L., & Ramírez-Alonso, G. (2022). Analysis of CNN architectures for human action recognition in video. Computación y Sistemas, 26(2), 623–641.
41. Wang, Y., Sun, Y., Li, Y., & Zhou, C. (2025). Malicious Document Detection Based on GGE Visualization. Computers, Materials and Continua, 82(1), 1233–1254. https://doi.org/10.32604/cmc.2024.057710
42. Xu, W., Nie, L., Chen, B., & Ding, W. (2023). Dual-stream EfficientNet with adversarial sample augmentation for COVID-19 computer aided diagnosis. Computers in Biology and Medicine, 165, 107451. https://doi.org/10.1016/j.compbiomed.2023.107451
43. Yao, L., Liu, B., & Xin, Y. (2024). Visualization-based comprehensive feature representation with improved EfficientNet for malicious file and variant recognition. Journal of Information Security and Applications, 86, 103865. https://doi.org/10.1016/j.jisa.2024.103865
44. Yu, Q., Zhang, Y., Xu, J., Zhao, Y., & Zhou, Y. (2024). Intelligent damage classification for tensile membrane structure based on continuous wavelet transform and improved ResNet50. Measurement, 227, 114260. https://doi.org/10.1016/j.measurement.2024.114260
45. Zhou, G., He, Q., Liu, X., Kai, X., Cao, W., Ding, J., Zhuang, B., Xu, S., & Thwin, M. (2024). Optimizing MobileNetV2 for improved accuracy in early gastric cancer detection based on dynamic pelican optimizer. Heliyon, 10(16), e35854. https://doi.org/10.1016/j.heliyon.2024.e35854
46. Zou, Y., Wu, L., Zuo, C., Chen, L., Zhou, B., & Zhang, H. (2025). White blood cell classification network using MobileNetv2 with multiscale feature extraction module and attention mechanism. Biomedical Signal Processing and Control, 99, 106820. https://doi.org/10.1016/j.bspc.2024.106820