Faster-YOLOv6: A Lightweight Target Detection Algorithm Optimized for Edge Computing

With the rapid advancement of edge computing, deploying target detection algorithms on resource-constrained devices has become a key challenge in computer vision. To address this, we propose Faster-YOLOv6, a lightweight target detection algorithm based on YOLOv6. First, we introduce a Partially Reparameterized Convolutional Module (PRepConv), an enhancement of the partially convolutional module (PConv) using reparameterization. This lightweight module improves feature extraction while reducing computational cost. Next, we design Faster RepBlock, a lightweight feature extraction module that enhances cross-channel information fusion and feature interaction, optimizing the network's efficiency. We replace the original RepBlock in YOLOv6 with Faster RepBlock, reducing redundant computations and memory access. Our experiments on a custom dataset deployed on a Jetson Nano show that Faster-YOLOv6 achieves a 40% faster inference speed compared to the baseline model, with no loss in accuracy. The improved model outperforms other mainstream detection algorithms in terms of speed and efficiency for edge computing.