Application of YOLOv4 Algorithm for Foreign Object Detection on a Belt Conveyor in a Low-Illumination Environment

Area of Science:

• Industrial Engineering
• Computer Vision
• Machine Learning

Background:

• Belt conveyors are prone to failures like runout, coal piles, and longitudinal tears.
• Current detection methods for longitudinal tearing are insufficient.
• Machine learning and machine vision offer potential for early detection of foreign objects, a key factor in preventing tears.

Purpose of the Study:

• To develop and validate a real-time machine vision system for detecting foreign items on belt conveyors.
• To enhance the effectiveness of early-stage detection for longitudinal belt tears.

Main Methods:

• Utilized the KinD++ algorithm for low-light image enhancement to improve image quality.
• Applied the GridMask method for data augmentation by partially masking foreign objects in training images.
• Integrated an optimized YOLOv4 algorithm for efficient foreign object detection.

Main Results:

• The proposed machine vision method successfully achieved real-time detection of foreign objects on belt conveyors.
• Image enhancement and data augmentation techniques improved the robustness of the detection system.
• The optimized YOLOv4 algorithm demonstrated high efficiency in identifying foreign items.

Conclusions:

• The developed machine vision approach, incorporating KinD++ and YOLOv4, is effective for detecting foreign objects on belt conveyors.
• This method provides a promising solution for minimizing longitudinal belt tears and improving conveyor safety.
• Further advancements in machine learning can enhance the reliability of automated inspection systems in industrial settings.