Year: 2024
Type: conference-paper
Abstract: The detection accuracy of tank targets in battlefield environments is often inadequate due to their diverse shapes and ability to blend into the surroundings. To address this issue, an improved YOLO v8 model is proposed for the precise detection of tank targets. First, Adaptive Kernel Convolution (AKConv) replaces the standard convolution in the YOLO v8 feature extraction layer to enhance adaptability to varying tank shapes. Second, a Mixed Local Channel Attention (MLCA) mechanism is integrated into the C2F module to strengthen feature extraction and improve detection accuracy. Third, the Shape IoU loss function substitutes the original loss function to accelerate model convergence. The improved model was trained, validated, and tested on a custom dataset. Results demonstrate that the improved YOLO v8 achieves 98.6% accuracy, 97.8% recall, and 99.3% mean average precision ([email protected]), surpassing the original YOLO v8 by 1.0%, 3.5%, and 1.1%, respectively, while reducing floating-point operations (GFLOPs) by 4.9%. This study offers an effective approach for accurate tank detection in battlefield environments.
Authors: Yugang Shan, Siyuan Liu, Jinquan Nie
Institution: Hubei University of Arts and Science
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4409829920