基于深度学习的废钢检测类别平衡策略和分组采样模块

Scrap Steel Detection Category Balancing Strategy and Group Sampling Module Based on Deep Learning

废钢的回收过程中,我们经常遇到种类繁多、类别数量不平衡等问题。鉴于此,本研究基于深度学习,提出了一种废钢检测方法,包括类别平衡策略(Class Balance)和分组采样模块(Multi Group Sampling)。类别平衡策略旨在解决数据集中存在的类别分布不均衡问题,而分组采样模块通过促进形状和大小相似的不同类别废钢之间的相互学习。通过对模型结构和训练流程的优化,该方法在废钢数据集上展现了出色的性能。我们采用rtmdet、yolov5和yolov8进行了一系列对比实验,结果显示本研究提出的策略能够在不同模型上取得更优的废钢图像检测效果,mAP分别提高了3.2%、2.6%和3.1%。本研究的成果为废钢回收处理行业提供了一种新的方案,提升废钢回收的效率和质量,推动废钢回收自动化的发展。

In the process of recycling scrap,we often encounter problems such as a wide variety of types and an imbalance in the number of categorics.In view of this,based on decp learning,this study proposes a scrap detection method,including a class balance strategy and a multi group sampling module.The category balancing strategy aims to solve the problem of uneven category distribution in the dataset,and the group sampling module facilitates mutual learning between different categories of scrap of similar shape and size.By optimizing the model structure and training process,the method in this study shows excellent performance on the scrap dataset.We conducted a series of comparative experiments using rtmdet,yolov5 and yolov8,and the results showed that the proposed strategy could achieve better scrap image detection results on different models,with mAP increased by 3.2%,2.6% and 3.1%,respcctively.The results of this study provide a new solution for the scrap rccycling industry,improve the efficiency and quality of scrap recycling,and promote the development of scrap recycling automation.