EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A THREE-CHANNEL NEURAL NETWORK DETECTION FRAMEWORK INTEGRATING SCENE, OBJECT, AND BEHAVIOR ANALYSIS

Jin Lu, Qin Zhou

Journal of Computer Science and Electrical Engineering · 2026

DOI: 10.61784/jcsee3158 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computer Science and Electrical Engineering, 2026. Verfasst von Jin Lu und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Three neural channels running side by side are the core of the detection framework presented here, and they let scene context, object recognition and behavior analysis be read together instead of one after another. Conventional pipelines either keep the three apart or chain them in a fixed order, and in both cases the dependencies among environmental context, object semantics and agent behavior drop out. Our pathways are the Scene Context Channel (SCC), which performs hierarchical spatial feature extraction; the Object Recognition Channel (ORC), built around a multiscale attention mechanism; and the Behavior Analysis Channel (BAC), which handles temporal dynamic modeling. A cross-attention fusion module joins them, learning adaptive weights that govern how much each channel draws on the others. Static images and video sequences are both accepted, and the three analysis dimensions are inferred jointly in real time. For validation we ran the method over several benchmark datasets covering surveillance, autonomous driving and human-computer interaction. Set against single-channel baselines and current multi-task architectures, it improves markedly on joint “scene-object-behavior” recognition and reaches state-of-the-art accuracy. Generalization holds across a range of environmental conditions, and the margin widens precisely where one would predict, namely in ambiguous scenes that no single modality can resolve. The ablations support the three-channel design as well, since each channel supplies information the others lack, and together they raise overall detection accuracy by 12-18% relative to isolated processing.

Zitieren

APA

Lu, J., & Zhou, Q. (2026). A THREE-CHANNEL NEURAL NETWORK DETECTION FRAMEWORK INTEGRATING SCENE, OBJECT, AND BEHAVIOR ANALYSIS. Journal of Computer Science and Electrical Engineering. https://doi.org/10.61784/jcsee3158

Vancouver

Lu J, Zhou Q. A THREE-CHANNEL NEURAL NETWORK DETECTION FRAMEWORK INTEGRATING SCENE, OBJECT, AND BEHAVIOR ANALYSIS. Journal of Computer Science and Electrical Engineering. 2026. doi:10.61784/jcsee3158

BibTeX
@article{Lu2026,
  author = {Lu, Jin and Zhou, Qin},
  title = {A THREE-CHANNEL NEURAL NETWORK DETECTION FRAMEWORK INTEGRATING SCENE, OBJECT, AND BEHAVIOR ANALYSIS},
  journal = {Journal of Computer Science and Electrical Engineering},
  year = {2026},
  doi = {10.61784/jcsee3158}
}
Kostenlos starten →