面向生活垃圾图像分类的多级特征加权融合算法

Multi-level features weighted fusion algorithm for domestic garbage image classification

生活垃圾的有效分类处理是改善社会环境的一项重要措施,传统的垃圾分类算法已经不能满足当前垃圾分类的要求。为了提高生活垃圾分类性能,在深入研究卷积神经网络中不同层次具有不同特征的基础上,提出一种面向生活垃圾图像分类的多级特征加权融合算法。构建基于ResNet的特征提取网络,通过多分支网络结构提取并处理图像不同层次的特征信息,在特征融合过程中分析自适应权重融合和固定权重融合对图像分类性能的影响,选取更优的加权融合方法对多级特征进行融合,从而获取更丰富的生活垃圾图像特征信息,提高垃圾分类准确率。实验结果表明,该算法在华为生活垃圾图像数据集上的分类准确率最高可达97.53%,优于其他算法,具有一定的实用价值。

The effective classification and treatment of domestic garbage is an important measure to improve the social environment. Traditional garbage classification algorithms can no longer meet the requirements of current garbage classification. In order to improve the classification performance of domestic waste, a multi-level feature weighted fusion algorithm for domestic waste image classification is proposed on the basis of in-depth study of the different characteristics of different layers in the convolutional neural network. A feature extraction network is construct based on ResNet, image feature information at different levels is extracted and processed through a multi-branch network structure, the impact of adaptive weight fusion and fixed weight fusion on image classification performance is analyzed to select a better weighted fusion during the feature fusion process.The method integrates multi-level features to obtain richer information on the image characteristics of domestic garbage and improves the accuracy of garbage classification. The experimental results show that the classification accuracy of this algorithm on the Huawei domestic garbage image data set can reach up to 97.53%, which is better than other algorithms, and has certain re-search and practical value.