PushPull-YOLO: A biologically inspired framework for robust object detection under image corruptions

Year 2025, Volume: 14 Issue: 3, 1100 - 1115, 15.07.2025

Hasan Ali Akyürek

https://doi.org/10.28948/ngumuh.1662465

Abstract

In this study, a novel integration of PushPull-Convolutional Layers into the YOLOv11 object detection model is proposed to enhance robustness against diverse image corruptions. The PushPull-Conv layer is designed based on biological mechanisms of the primary visual cortex, where complementary push and pull kernels are utilized to improve selectivity by amplifying relevant stimuli and suppressing irrelevant noise. The initial convolutional layer of YOLOv11 is replaced by this modification, and performance is evaluated on the COCO dataset across 15 corruption types (e.g., noise, blur, weather, digital artifacts) with five severity levels. Improved robustness metrics are achieved by the PushPull-enhanced YOLOv11 compared to the baseline. Detection performance under challenging conditions, including brightness variation, motion blur, and contrast changes, is enhanced. A link is established between biologically inspired design and deep learning, positioning PushPull-YOLO as a promising solution for real-time object detection in dynamic environments, with potential extensions to segmentation and keypoint detection.

Keywords

Convolutional Layers , Deep Learning , Image Corruption , Object Detection , PushPull , YOLO

PushPull-YOLO: Görüntü bozulmaları altında güçlü nesne algılama için biyoloji esinli bir çerçeve

Abstract

Bu çalışmada, YOLOv11 nesne tespit modeline PushPull Konvolüsyon Katmanlarının özgün bir entegrasyonu önerilerek görüntü bozulmalarına karşı dayanıklılığın artırılması amaçlanmıştır. PushPull-Conv katmanı, birincil görsel korteksin biyolojik mekanizmalarından esinlenerek tasarlanmış ve tamamlayıcı push ve pull çekirdekleri kullanılarak ilgili uyaranların güçlendirilmesi ve ilgisiz gürültünün bastırılması yoluyla seçiciliğin artırılması sağlanmıştır. YOLOv11’in ilk konvolüsyon katmanı bu değişiklik ile değiştirilmiş ve performans, COCO veri kümesi üzerinde 15 farklı bozulma türü (ör. gürültü, bulanıklık, hava koşulları ve dijital bozulmalar) ve beş şiddet düzeyinde değerlendirilmiştir. PushPull ile güçlendirilmiş YOLOv11’in, temel modele kıyasla üstün dayanıklılık metrikleri elde ettiği gösterilmiştir. Parlaklık değişimi, hareket bulanıklığı ve kontrast farklılıkları gibi zorlu koşullar altında tespit performansı iyileştirilmiştir. Biyolojik esinli tasarım ile derin öğrenme arasında bir bağlantı kurulmuş ve PushPull-YOLO’nun dinamik ortamlarda gerçek zamanlı nesne tespiti için umut verici bir çözüm sunduğu ortaya konulmuştur. Ayrıca yöntemin gelecekte segmentasyon ve anahtar nokta tespiti gibi diğer bilgisayarla görme görevlerine de uygulanabileceği düşünülmektedir.

Keywords

Evrişimsel Katmanlar , Derin Öğrenme , Görüntü Bozulması , Nesne Algılama , PushPull , YOLO

References

• Y. Dong, C. Kang, J. Zhang, Z. Zhu, Y. Wang, Y. Xiao, H. Su, X. Wei, and J. Zhu, Benchmarking robustness of 3d object detection to common corruptions in autonomous driving, arXiv Preprint, 2023. https://doi.org/10.48550/arxiv.2303.11040.
• H. A. Akyürek, H. İ. Kozan, and Ş. Taşdemir, Surface crack detection in historical buildings with deep learning-based YOLO algorithms: a comparative study, Computational Research Progress in Applied Science & Engineering, 10, 3, 1-14, 2024. https://doi.org/10.61186/crpase.10.3.2904.
• M. Akgül, H. İ. Kozan, H. A. Akyürek, and Ş. Taşdemir, Automated stenosis detection in coronary artery disease using yolov9c: Enhanced efficiency and accuracy in real-time applications, Journal of Real-Time Image Processing, 21, 177, 2024. https://doi.org/10.1007/s11554-024-01558-x.
• H. A. Akyürek, Tıbbi Görüntülerde Yapay Zekâ Temelli Nesne Tespiti Uygulamaları, in Bilgisayar Bilimleri ve Mühendisliğinde İleri Araştırmalar, G. Kutluana Ed. Afyonkarahisar: YAZ Yayınları, 2024. https://doi.org/10.5281/zenodo.14604344.
• H. İ. Kozan and H. A. Akyürek, Efficient and rapid classification of various maize seeds using transfer learning and advanced AI techniques, ASEAN Journal of Scientific and Technological Reports, 28, 1, e255200, 2025. https://doi.org/10.55164/ajstr.v28i1.255200.
• J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, You only look once: unified, real-time object detection, IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 779-788, 2016. https://doi.org/10.1109/cvpr.2016.91.
• R. Kishor, Performance benchmarking of YOLOv11 variants for real-time delivery vehicle detection: a study on accuracy, speed, and computational trade-offs, Asian Journal of Research in Computer Science, 17, 12, 108-122, 2024. https://doi.org/10.9734/ajrcos/2024/v17i12532.
• R. Khanam and M. Hussain, YOLOv11: an overview of the key architectural enhancements, arXiv Preprint, 2024. https://doi.org/10.48550/arXiv.2410.17725.
• H. Hsiang, Development of an effective corruption-related scenario-based testing approach for robustness verification and enhancement of perception systems in autonomous driving, Sensors, 24, 1, 301, 2024. https://doi.org/10.3390/s24010301.
• C. Michaelis, B. Mitzkus, R. Geirhos, E. Rusak, O. Bringmann, A. S. Ecker, M. Bethge, and W. Brendel, Benchmarking robustness in object detection: autonomous driving when winter is coming, arXiv Preprint, 2020. https://doi.org/10.48550/arXiv.1907.07484.
• G. S. Bennabhaktula, E. Alegre, N. Strisciuglio, and G. Azzopardi, PushPull-Net: inhibition-driven ResNet robust to image corruptions, in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2025, 15308 LNCS, 391-408, https://doi.org/10.1007/978-3-031-78186-5_26.
• T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollár and C. L. Zitnick, Microsoft COCO: common objects in context, Cham, 2014: Springer International Publishing, in Computer Vision – ECCV 2014, 740-755. https://doi.org/10.1007/978-3-319-10602-1_48.
• P. Varghese, Biologically inspired deep residual networks, Iaes International Journal of Artificial Intelligence (Ij-Ai), vol. 12, no. 4, p. 1873, 2023. https://doi.org/10.11591/ijai.v12.i4.pp1873-1882.
• R. Pogodin, Y. Mehta, T. P. Lillicrap, and P. E. Latham, Towards biologically plausible convolutional networks, 2021. https://doi.org/10.48550/arxiv.2106.13031.
• Y. Shen, J. Wang, and S. Navlakha, A correspondence between normalization strategies in artificial and biological neural networks, Neural Computation, 33 ,12 3179–3203, 2021. https://doi.org/10.1162/neco_a_01439.
• M. Zingerenko, Bipolar morphological YOLO network for object detection, Proceedings Volume 13072, Sixteenth International Conference on Machine Vision 130720Q, 2024. https://doi.org/10.1117/12.3023255.
• D. H. Hubel and T. N. Wiesel, Receptive fields, binocular interaction and functional architecture in the cat's visual cortex, The Journal of Physiology, 160, 1, 106, 1962. https://doi.org/10.1113/jphysiol.1962.sp006837.
• G. Azzopardi, A. Rodríguez-Sánchez, J. Piater, and N. Petkov, A push-pull CORF model of a simple cell with antiphase inhibition improves SNR and contour detection, PloS one, 9, 7, e98424, 2014. https://doi.org/10.1371/journal.pone.0098424.
• L. J. Borg-Graham, C. Monier, and Y. Fregnac, Visual input evokes transient and strong shunting inhibition in visual cortical neurons, Nature, 393, 6683, 369-373, 1998. https://doi.org/10.1038/30735.
• D. H. Hubel and T. N. Wiesel, Brain mechanisms of vision, Scientific American, 241, 3, 150-163, 1979. https://doi.org/10.1038/scientificamerican0979-150.
• H. Zhang and Y. Wang, Towards adversarially robust object detection, 2019 IEEE/CVF International Conference on Computer Vision, Seoul, Korea, pp. 421-430, 2019. https://doi.org/10.1109/iccv.2019.00051.
• R. Poojary, R. Raina, and A. K. Mondal, Effect of data-augmentation on fine-tuned cnn model performance, Iaes International Journal of Artificial Intelligence, 10, 84, 2021. https://doi.org/10.11591/ijai.v10.i1.pp84-92.
• P. Shah and M. El-Sharkawy, A-MnasNet: augmented MnasNet for computer vision, 2020 IEEE 63rd International Midwest Symposium on Circuits and Systems (MWSCAS), pp. 1044-1047, Springfield, MA, USA, 2020. https://doi.org/10.1109/mwscas48704.2020.9184619.
• C. Shorten and T. M. Khoshgoftaar, A survey on image data augmentation for deep learning, Journal of Big Data, 6, 1, 2019. https://doi.org/10.1186/s40537-019-0197-0.
• Ultralytics. Ultralytics YOLO documentation. https://docs.ultralytics.com/models/yolo11/, Accessed 20 March 2025.
• Y. Luo, The evolution of YOLO: from YOLOv1 to YOLOv11 with a focus on YOLOv7's innovations in object detection, Theoretical and Natural Science, 87, 1, 82-90, 2025. https://doi.org/10.54254/2753-8818/2025.20335.
• E. H. Alkhammash, A comparative analysis of YOLOv9, YOLOv10, YOLOv11 for smoke and fire detection, Fire, 8, 1, 26, 2025. https://doi.org/10.3390/fire8010026.
• R. Bohush, S. Ablameyko, and Y. Adamovskiy, Robust object detection in images corrupted by impulse noise, Computer Modeling and Intelligent Systems, vol. 2608, pp. 1107-1116, 2020. https://doi.org/10.32782/cmis/2608-83.
• B. Jaison, G. Anjali Jha, J. Jeevitha, and C. Devi Priya, You only look once(YOLO) object detection with COCO using machine learning, in International Interdisciplinary Humanitarian Conference for Sustainability, IIHC 2022 - Proceedings, 2022. pp. 1574-1578, https://doi.org/10.1109/IIHC55949.2022.10059737.
• A. Borji, Complementary datasets to COCO for object detection, arXiv Preprint, 2022. https://doi.org/10.48550/arxiv.2206.11473.
• R. F. Woolson, Wilcoxon signed‐rank test, Encyclopedia of Biostatistics, 8, 2005. https://doi.org/10.1002/0470011815.b2a15177.
• C. Lu, J. Wu, S. Ali, and M. L. Olsen, Assessing the uncertainty and robustness of object detection models for detecting stickers on laptops, arXiv Preprint, 2024. https://doi.org/10.48550/arXiv.2409.03782.