Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (23): 293-299.DOI: 10.3778/j.issn.1002-8331.2105-0443

• Engineering and Applications • Previous Articles     Next Articles

Recyclable Garbage Detection Method of Improved SSD

GENG Liting, Alifu·Kuerban, Minawaer·Abula, DING Pei, JIANG Runxi   

  1. School of Software, Xinjiang University, Urumqi 830046, China
  • Online:2022-12-01 Published:2022-12-01



  1. 新疆大学 软件学院,乌鲁木齐 830046

Abstract: Nowadays, the recycling and utilization of recyclable garbage in our country mainly relies on manual sorting, which has caused problems such as waste of human resources and low resource utilization. To improve the recycling of resources, a recyclable garbage detector based on improved single shot multibox detector(SSD) algorithm is proposed. In order to solve the problem that the model parameters are too large to deploy, a new feature extraction backbone is selected, a lightweight network RepVGG is introduced to replace the VGG16 network in SSD, and structural re-parameterization method is used to significantly reduce the parameters and computation. It modifies the auxiliary convolution layer structure of SSD to further reduce parameters. For the large change of dataset size, SK module is introduced to adjust the size of the receptive field adaptively and improve the detection accuracy. The experimental results show that the improved SSD model has better detection accuracy and real-time performance in the detection task of recyclable garbage, the accuracy is 95.23%, which is 4.33?percentage points higher than the original SSD, and the detection speed can reach 64?FPS, so the algorithm can be better applied in industry.

Key words: recyclable garbage detection, single shot multibox detector(SSD), structural re-parameterization, deep learning

摘要: 目前我国可回收垃圾的回收利用主要依靠人工分拣的方式,造成了人力资源浪费、资源利用率低等问题。为提高资源的回收利用,提出了一个基于改进single shot multibox detector(SSD)算法的可回收垃圾检测器。针对模型参数量大,难以部署应用的问题,选用新的主干特征提取网络,引入轻量化的网络RepVGG替换SSD中的VGG16网络,同时采用结构重参数化的方式大幅减少参数量和计算量。修改SSD的辅助卷积层结构,进一步减少参数量。针对数据集尺寸变化大的问题,引入SK模块,自适应调整感受野的尺寸,提高检测精度。实验结果表明,改进的SSD模型在可回收垃圾检测任务上具有较好的检测精度和实时性,精度为95.23%,相较于原始SSD提升了4.33个百分点,检测速度可以达到64?FPS,因此该算法可以更好地应用于工业。

关键词: 可回收垃圾检测, SSD算法, 结构重参数化, 深度学习