A Lightweight Object Detection Algorithm Based on Improved YOLOv8

Abstract

Lightweight object detection algorithms are crucial in the field of computer vision, directly affecting whether computer vision algorithms can be deployed on resource-constrained devices and meet the real-time requirements of daily life. To address the above problems, this paper proposes a lightweight object detection algorithm LAD-YOLO based on improved YOLOv8. First, we optimize the point-wise convolution in depthwise separable convolution to enhance the model's learning ability, introduce depthwise separable convolution into the backbone network and neck network to reduce the model size, and construct a lightweight detection head. Meanwhile, the LSKA (Large Separable Kernel Attention) mechanism is introduced to help the model capture multi-scale information and achieve better detection performance. Extensive experiments conducted on the VOC dataset show that the proposed LAD-YOLO algorithm improves the precision (P) and mAP0.5:0.95 by 2.5% and 1.8% respectively compared with YOLOv8n, while maintaining lower parameters and computational complexity.