摘要
针对生活垃圾分类中存在的样本不平衡问题,本文基于实例分割模型Mask R-CNN神经网络展开研究,通过调整神经网络的损失函数和扩充数据集中的难分正样本来提升神经网络的检测精度。实验结果表明,该方法降低了数据不平衡对网络性能的影响,使生活垃圾的回收率得到了提升。
In response to the issue of sample imbalance in the classification of household waste,this study explores the Mask R-CNN neural network model for instance segmentation.By adjusting the neural network's loss function and augmenting difficult-to-separate positive samples in the dataset,the detection accuracy of the neural network is improved.Experimental results indicate that the approach has reduced the impact of data imbalance on network performance,leading to an improvement in the recycling rate of household waste.
作者
王智峰
WANG Zhifeng(Xiamen Luhai Pro-environment Inc.,Xiamen,China,361001)
出处
《福建电脑》
2024年第2期22-26,共5页
Journal of Fujian Computer
关键词
生活垃圾分选
深度学习
样本不平衡
损失函数
Municipal Solid Waste Sorting
Deep Learning
Sample Imbalance
Loss Function
