GVCG-YOLO: A fast and lightweight Twin-fusion fiber-core algorithm for efficient kitchen waste detection

• A multi-label mixed MKTrash dataset is constructed. • A GVCG-YOLO model of real-time kitchen waste detection is proposed. • An intelligent kitchen-waste detection and sorting system is designed. • The proposed model detected the kitchen waste with 95.1% mAP and 5.5 GFLOPs. With rising living standards and accelerated urbanization, the amount of kitchen waste is increasing significantly. Traditional manual sorting methods are inefficient and costly, making kitchen waste management a significant challenge for sustainable urban development. To enhance processing efficiency and accuracy, we propose a rapid and lightweight GVCG-YOLO model for effective kitchen waste detection. In this study, we introduce the Twin-Fusion Fiber-Core module, which integrates GSConv and VoV-GSCSP. GSConv generates ghost feature maps to expand feature representation without increasing computational load significantly, improving the recognition of small or partially occluded kitchen waste in complex backgrounds. VoV-GSCSP optimizes gradient propagation through staged weight sharing and residual connections, enhancing deep feature learning and improving recognition accuracy across various kitchen waste types. Additionally, we incorporate the C3Ghost module into the enhanced YOLOv8n backbone network. C3Ghost combines the CSP architecture and Ghost module to reduce parameters and computational burden while enhancing the recognition of subtle kitchen waste features in complex backgrounds. Its lightweight operation improves feature reuse, ensuring sufficient information flow in the deep network. This approach not only achieves accurate positioning and recognition of kitchen waste objects but also significantly improves processing speed and real-time performance. The intelligent kitchen waste sorting system designed by our team optimizes image acquisition and data processing through precise circuit layout and power management strategies. This results in a notable improvement in system response speed and stability, providing robust support for recognizing and sorting kitchen waste in complex environments, and contributing to the advancement of intelligent waste management.