LWCNet: A Lightweight and Efficient Algorithm for Household Waste Detection and Classification Based on Deep Learning

The existing waste classification algorithms based on computer vision are hindered by issues such as low classification accuracy, a large number of model parameters, and high computational complexity. Aiming at these problems, a Lightweight Waste Classification Network (LWCNet) based on the basic framework of YOLOv8n is designed and proposed in this paper. Firstly, a detection head based on the self-attention mechanism, SAHead, is proposed to capture a wider range of contextual information. Secondly, the lightweight convolution GSConv is employed to optimize the standard convolution in YOLOv8n. Combined with GELAN, the GRCSPELAN module is designed to reduce the model size without compromising performance. Finally, the AIFI module is introduced to enhance the intra-scale feature interaction ability. The experimental results indicate that LWCNet effectively reduces the model size and enhances detection efficiency, providing reliable theoretical support for the automation of household waste detection and classification.