【摘要】 随着全球环保意识的提高和资源回收利用的重要性日益凸显，对可回收垃圾的准确识别成为一项关键任务。提出了一种基于注意力机制的可回收垃圾识别方法，该方法使用Mask R-CNN网络模型来提取垃圾图像中的纹理、形状和颜色等特征。同时，引入了ECA注意力机制，通过学习注意力权重，模型能够更有针对性地关注那些与垃圾分类最为紧密相关的特征。此外，还使用了GIoU损失函数来更好地反映边界框之间的相对位置和大小关系。经实验验证，改进后的模型m AP达到了81.7%，相比原模型提升了8.5%。更多还原

【Abstract】 As global environmental awareness rose and the importance of resource recycling became increasingly prominent,the accurate identification of recyclable waste is recognized as a crucial task. A method for recyclable waste identification based on the attention mechanism was proposed. This method utilized the Mask R-CNN network model to extract features such as texture, shape, and color from waste images. At the same time, the ECA attention mechanism was introduced, allowing the model, through learning attention weights, to focus more selectively on those features most closely related to waste classification.Additionally, the GIoU loss function was employed to better reflect the relative position and size relationships between bounding boxes. Experimental verification showed that the improved model achieved an mAP of 81.7%, representing an 8.5% increase compared to the original model.更多还原