YOLO-STOD: an industrial conveyor belt tear detection model based on Yolov5 algorithm

Area of Science:

• * Computer Vision
• * Deep Learning
• * Industrial Safety

Background:

• * Real-time detection of conveyor belt tears is critical for safety and operational efficiency in the coal industry.
• * Existing methods struggle with the multi-scale nature, small targets, and complex interference characteristic of conveyor belt tears.
• * There is a need for improved algorithms capable of robustly detecting small-size tear damage under challenging industrial conditions.

Purpose of the Study:

• * To develop an advanced deep learning-based visual detection method for real-time identification of conveyor belt tears.
• * To address the challenges of multi-scale targets, abundant small targets, and complex interference in tear detection.
• * To enhance the performance and robustness of conveyor belt tear detection systems for industrial applications.

Main Methods:

• * Development of a specialized multi-case conveyor belt tear dataset to accommodate complex interference and small-size detection scenarios.
• * Design of the YOLO-STOD detection method, incorporating the BotNet attention mechanism for enhanced feature extraction, particularly for small targets.
• * Utilization of Shape_IOU for calculating training loss, considering bounding box shape regression to improve model robustness.

Main Results:

• * The YOLO-STOD method demonstrated superior performance compared to competing methods in experimental evaluations.
• * Achieved a recall of 91.2%, a Map value of 91.9%, and a Frames Per Second (FPS) of 190.966.
• * The method effectively addresses the challenges of small targets and complex interference in conveyor belt tear detection.

Conclusions:

• * The proposed YOLO-STOD visual detection method is highly effective for real-time conveyor belt tear detection.
• * The system's performance metrics satisfy the stringent requirements for industrial real-time detection.
• * YOLO-STOD is a promising solution for enhancing safety and operational monitoring in the industrial mining sector.