一种基于YOLOv5的家用物体检测优化算法

An Optimization Algorithm for Household Object Detection Based on YOLOv5

针对现有家用物体检测算法模型存在的计算量大和对小目标检测效果不佳等问题,本文提出了一种基于YOLOv5的家用物体检测优化算法。采用轻量级的GhostBottleneck,代替Bottleneck结构,降低网络参数。同时,添加卷积注意力模块,强化小目标物体的特征信息,从而提高对家用小物体的检测性能。为了证明该算法的有效性,采用YOLOv5m_G、YOLOv5m_GC_a和YOLOv5m_GC_b 3种网络模型,在自建的家用物体数据集的训练集和验证集中进行训练,并在测试集上对模型的性能进行对比分析。研究结果表明,在保证检测精度的前提下,改进后的算法YOLOv5m_GC_b与原始的YOLOv5m算法相比,参数量降低了30%,计算量降低了37%,有效降低了参数量和计算量,便于更好地部署在嵌入式设备中,提高了对家用小物体的检测性能,该研究具有一定的创新性。

Aiming at the problems of the existing household object detection algorithm model,such as large amount of computation and poor detection effect on small objects,this paper proposes an optimization algorithm for household object detection based on YOLOv5.It uses lightweight GhostBottleneck to replace the Bottleneck structure and reduces network parameters.At the same time,the convolution attention module is added to the algorithm to strengthen the feature information of small target objects,so as to improve the detection performance of household small objects.In order to prove the effectiveness of the algorithm,three network models,YOLOv5m_G,YOLOv5m_GC_a and YOLOv5m_GC_b,are proposed in this paper.The model is trained in the training set and the validation set of the self-built household object dataset,and its performance is compared and analyzed on the test set.The research results show that,compared with the original YOLOv5m algorithm,the improved algorithm YOLOv5m_GC_b reduces the number of parameters by 30%and the computation amount by 37%under the premise of ensuring the detection accuracy.It can effectively reduce the number of parameters and the amount of computation,and facilitate better deployment in embedded devices.This research is innovative.