Real time wire rope detection method based on Rockchip RK3588
Abstract In the era of fully mechanized and automated coal mine production, the need for autonomous fault detection and intelligent identification of wire ropes has become increasingly critical. Conventional model training and algorithmic computations rely on server-based systems, requiring signific...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-08-01
|
| Series: | Scientific Reports |
| Online Access: | https://doi.org/10.1038/s41598-025-16043-z |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract In the era of fully mechanized and automated coal mine production, the need for autonomous fault detection and intelligent identification of wire ropes has become increasingly critical. Conventional model training and algorithmic computations rely on server-based systems, requiring significant computational resources. This study proposes a real-time wire rope detection system utilizing the Rockchip RK3588 platform. To enhance non-destructive wire rope inspection, a Mini-YOLO model was developed by integrating MobileNetV3, the Coordinate Attention (CA) mechanism, and a novel loss function, Inner-IoU, into the YOLOv8 framework. This paper’s innovation lies not in creating algorithmic components from scratch, but in their synergistic integration and targeted optimization to solve the specific challenges of real-time defect detection on resource-constrained edge devices. To optimize the Neural Network Processing Unit (NPU) for computational performance, a thread pool is implemented with the C++ programming language to partition and accelerate the output processing of the model. Additionally, a Docker container is employed for environment configuration, simplifying deployment. After testing, the raw data and test results are stored in real-time, with periodic uploads to the cloud for data backup. Experimental findings demonstrate that the Mini-YOLO algorithm achieves a computational speed 2 times faster than YOLOv8, with an accuracy improvement of 1.2%. After deployment, the inference time per image is approximately 18.5 ms per image, enabling efficient real-time detection. |
|---|---|
| ISSN: | 2045-2322 |