Research and Development of Waste Sorting System Based on Improved YOLOv5 Algorithm and Attention Mechanism in Edge Environment

To address the limitations of traditional garbage sorting methods that rely on manual intervention, this paper presents an innovative intelligent waste classification system that combines deep learning-based image recognition technology with automated robotic arm control for the efficient identification and classification of four typical types of household waste. The core innovation of the system lies in the use of an improved YOLOv5 model, employing a lightweight MobileNet as the backbone network to achieve efficient feature extraction and incorporating an attention mechanism to enhance classification accuracy. By assigning greater weight to important features, the model significantly improves its ability to capture key waste characteristics, thereby increasing recognition accuracy. In terms of hardware selection, after evaluating various modules, the K210 module was ultimately chosen due to its outstanding advantages in power consumption, control capabilities, and intelligence performance. The control system is centered around the STM32F103C8T6 microcontroller, which centrally manages the waste recognition module, robotic arm, power supply, and communication modules, enabling real-time waste recognition and classification while coordinating the robotic arm and vehicles for efficient waste processing. The The system also developed the mobile terminal software enhancing the user interaction experience. Testing results indicate that this intelligent waste classification system demonstrates high intelligence, accurate waste classification capabilities, and good real-time operational performance along with modular expandability. Overall, this paper provides a new solution for intelligent waste classification by combining deep learning and automation technologies, promoting development in related fields, and showcasing the broad potential of deep learning in practical applications.