%0 Journal Article %T SWD-YOLO：基于YOLOv11n的高精度轻量级生活垃圾分类算法
SWD-YOLO: A High-Precision Lightweight Household Waste Classification Algorithm Based on YOLOv11n %A 谷晓琳 %A 申奥 %J Software Engineering and Applications %P 571-584 %@ 2325-2278 %D 2025 %I Hans Publishing %R 10.12677/sea.2025.143050 %X 近年来，垃圾分类已成为国际社会关注的热点，很多国家通过科技创新来推进垃圾分类工作。有效的垃圾分类，可以提高垃圾的回收率，保护环境，节约资源。为了有效地实现垃圾分类，本研究设计并开发了一种高精度且轻量级的生活垃圾分类目标检测模型。本文的技术创新集中在基于YOLOv11n算法引入的SPPF-LSKA模块，WaveletPool模块和DynamicConv模块三个模块的关键改进上。首先，通过引进SPPF-LSKA模块，使改进后的YOLOv11n算法显著提高了SPPF模块在多尺度上聚合特征的能力；其次，我们引入了WaveletPool模块，将小波滤波器应用于输入特征图，并将其输出结果沿着通道维度连接起来，进一步降低了冗余，提升了模型的识别精度；最后，我们引入DynamicConv模块进行优化，它通过动态聚合多个卷积核来增强模型表示能力，并且可以轻松集成到现有的CNN架构中。SWD-YOLO能够以较低的计算成本保持其优良的性能，从而显著提高模型识别效率。实验结果表明，与YOLOv11n相比SWD-YOLO的参数量几乎不变，计算量降低了14%，平均精度(mAP@50)提高了2.4%。改进后的垃圾分类检测模型不仅提高了平均精度，而且实现了模型的轻量级，从而保证了较高的操作效率。
In recent years, garbage classification has become a hot topic of concern for the international community, and many countries have promoted garbage classification through scientific and technological innovation. Proper garbage classification can improve the recycling rate of garbage, protect the environment and save resources. In order to realize garbage classification effectively, this study designs and develops a high-precision and lightweight object detection model for domestic garbage classification. The technological innovation of this paper focuses on the key improvements of three modules, SPPF-LSKA module, WaveletPool module and DynamicConv module, which are introduced based on the YOLOv11n algorithm. First, by introducing the SPPF-LSKA module, the improved YOLOv11n algorithm significantly improves the ability of the SPPF module to aggregate features at multiple scales; second, we introduce the WaveletPool module, which applies wavelet filters to the input feature maps and concatenates the results along the channel dimensions, which further reduces the redundancy and improves the recognition accuracy of the model; Finally, we introduce the DynamicConv module for optimization, which enhances the model representation by dynamically aggregating multiple convolutional kernels and can be easily integrated into existing CNN architectures. SWD-YOLO is able to maintain its excellent performance at a low computational cost, which significantly improves the model recognition efficiency. Experimental results show that SWD-YOLO has almost the same number of parameters compared to YOLOv11n, the computational effort is reduced by 14%, and the average accuracy (mAP@50) is improved by 2.4%. The improved garbage classification detection model not only improves the average accuracy, but also realizes the lightweight of the model, which ensures high %K YOLOv11n， %K 垃圾分类， %K 目标检测， %K mAP@50
YOLOv11n %K Garbage Classification %K Object Detection %K mAP@50 %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=117456