Artificial Intelligence in Automotive Industry
Artificial Intelligence · 30.06.2026 · 31 (AI.2026.31(2)) · 203-211
DOI: 10.15407/jai2026.02.203 ↗Kurz gefasst
Fachartikel in der Zeitschrift Artificial Intelligence, erschienen am 30.06.2026. Verfasst von Mamuta M und 2 weiteren Autor:innen. Die Arbeit nennt 31 Literaturangaben.
Abstract (Originalsprache)
Nowadays, artificial intelligence plays a crucial role in the automotive industry. It is used as machine learning, deep learning, neural networks, natural language processing, computer vision, fuzzy logic and other techniques to improve efficiency, speed and comfort. More and more popularity gains YOLO in the automotive industry that demonstrates outstanding performance and accuracy. However, efficiency of the latest versions, namely YOLOv11 and YOLO26, is underexplored. This article addresses this gap. Moreover, the influence of layer freezing technique on YOLOv11 and YOLO26 during detection and segmentation of different car parts is investigated, that is very essential in applications such as safety analysis, damage assessment, insurance and automated production. Experiments were conducted in Google Colaboratory Pro environment with GPU NVIDIA A100 (40Gb). Carparts Segmentation Dataset was used for experiments, containing 3833 labeled images split into three subsets: training, validation and test. Models were trained for 100 epochs with an early stopping mechanism (patience 20) to avoid overfitting, batch size 16, optimizer AdamW. It was found empirically that YOLOv11 outperform previous versions of YOLO series according to such metrics as Recall, F1 and mAP50-95. YOLOv11 without layer freezing is the best in terms of the F1 and mAP50-95 metrics. Layer freezing is beneficial for YOLOv11 that confirm such metrics as recall and mAP50. The best strategy is to freeze first 6 layers of the YOLOv11 architecture for its performance improvement. The latest version YOLO26 without layer freezing does not achieve better results than the previous versions according all metrics. Layer freezing does not show any improvement of YOLO26 model performance for all metrics. Results revealed that YOLOv11 with 6 layers frozen detects with higher confidence such car parts as “back light”, “back right door”, “back right light”, “front door” and “wheel” in comparison with YOLO26. In the future, it is planned to conduct a study of the models’ effectiveness for individual car parts, analyze the correlation between standard metrics and changes in the detection and identification of parts, and evaluate the models’ performance on images from the test subset.
Zitieren
APA
M, M., I, K., & O, M. (2026). Artificial Intelligence in Automotive Industry. Artificial Intelligence, 31(AI.2026.31(2)), 203-211. https://doi.org/10.15407/jai2026.02.203
Vancouver
M M, I K, O M. Artificial Intelligence in Automotive Industry. Artificial Intelligence. 2026;31(AI.2026.31(2)):203-211. doi:10.15407/jai2026.02.203
BibTeX
@article{M2026,
author = {M, Mamuta and I, Kravchenko and O, Mamuta},
title = {Artificial Intelligence in Automotive Industry},
journal = {Artificial Intelligence},
year = {2026},
volume = {31},
number = {AI.2026.31(2)},
pages = {203--211},
doi = {10.15407/jai2026.02.203}
}
EUVIMED