Authors:
K. Aravinda, V. Revathi, Thirumalraj Karthikeyan, S. Venkatasubramanian, Mykhailo Paslavskyi

Addresses:
Department of Electronics and Communication Engineering, New Horizon College of Engineering, Bengaluru, Karnataka, India. Department of Research and Development, New Horizon College of Engineering, Bengaluru, Karnataka, India. Department of Artificial Intelligence, Trichy Research Labs, Quest Technologies, Tiruchirappalli, Tamil Nadu, India. Department of Computer Science and Business Systems, Saranathan College of Engineering, Trichy, Tamil Nadu, India. Department of Computer Science, Ukrainian National Forestry University, Lviv, Lviv Oblast, Ukraine.

Abstract:

Crime rates are rising worldwide, making automated crime detection systems more vital. This study uses Fast Local Laplacian Filter FLLF preprocessing to improve the UCF-Crime dataset photos for automated crime identification. This solution preserves sharp edges and visual features needed to identify crimes. A novel interactive attention strategy is employed to extract features from the Visual Transformer (ViT) and Res2Net, collectively referred to as ViTRes2DualNet, leveraging their combined capabilities. The Adaptive Hybrid Attention Network (AdaHybridANet) is used for categorisation to maximise crime-detection accuracy. The proposed classifier enhances feature extraction by integrating the Coordinate Attention and Enhanced Non-Local Attention (ENLA) modules to extract both local and global features. The suggested model's hyperparameter tuning utilises the Lyrebird Optimisation Algorithm (LBOA) to simulate Lyrebirds in danger. LBOA replicates escape and hiding by mathematically modelling Lyrebirds' threat scanning and decision-making. Results revealed 99.32% accuracy for the suggested strategy. When utilising innovative AdaHybridANet classification and LBOA tuning, the proposed method outperforms existing methods. With less information wasted, this approach delivers high-quality images for more accurate detection.

Keywords: Contrast Enhancement; Fast Local Laplacian Filter; Convolutional Neural Network; Vision Transformer; Lyrebird Optimisation Algorithm; Enhanced Non-Local Attention; Adaptive Hybrid Attention Network.

Received on: 03/01/2025, Revised on: 12/03/2025, Accepted on: 25/04/2025, Published on: 22/11/2025

DOI: 10.69888/FTSCL.2025.000485

FMDB Transactions on Sustainable Computer Letters, 2025 Vol. 3 No. 4, Pages: 183-197

• Views : 25
• Downloads : 9

Download PDF