YOLOv7是一种目标检测算法,但其在一些资源受限的设备上可能面临计算和内存压力。为了解决该问题,本文提出了使用GSConv对YOLOv7模型进行改进,在减轻复杂度的同时,添加P2检测层和WIoU损失函数增加模型的检测性能。为了验证我们提出的方法的有效性,我们进行了一系列实验和比较。我们选取了标准的YOLOv7模型作为基准,并使用PASCAL VOC2007公开数据集进行训练和测试。实验结果表明,相较于标准模型,本文提出的模型可以显著减少参数量,同时保持较高的检测精度,为在资源受限的设备上部署和运行YOLOv7模型提供了可行的解决方案。YOLOv7 is an object detection algorithm, but it may encounter computational and memory pressures on certain resource-constrained devices. To tackle this challenge, this paper proposes enhancing the YOLOv7 model using GSConv, incorporating a P2 detection layer and a WIoU loss function to bolster the model’s detection capabilities while minimizing complexity. To validate the efficacy of our approach, we conducted a series of experiments and comparisons. We adopted the standard YOLOv7 model as the baseline and trained and tested it using the PASCAL VOC2007 public dataset. The experimental results demonstrate that, compared to the standard model, our proposed model can significantly reduce the number of parameters while maintaining high detection accuracy, offering a viable solution for deploying and running the YOLOv7 model on resource-limited devices.
