基于ResNeXt和注意力机制的垃圾分类算法

In this paper, ResNeXt is the baseline of the garbage classification algorithm. Based on ResNeXt, this paper tries to develop a new garbage classification algorithm. This paper uses the dataset of Huawei Cloud "Garbage Classification Competition" and carries out a number of data preprocessing work on the data set, including the adjustment of the image size, standardization and data augmentation, which consists of several methods including spatial geometric transformation, random erasing and Mixup. This paper verifies the result of the data augmentation, which shows that data augmentation plays a signification role in improving the classification result. Also, Attention Module is involved in improving the classification result. Attention Module consists of Channel Attention Module and Space Attention Module. The experiment result shows that adding an combination of Channel Attention Module and Space Attention Module before and after the ResNeXt is the best solution. To sum up, this paper proposes a garbage classification model based on ResNeXt and adding Attention Module before and after the network.