Research on multi dimensional feature extraction and recognition of industrial and mining solid waste images based on mask R-CNN and graph convolutional networks
Abstract Aiming at the problems of traditional methods for multi-dimensional feature extraction of industrial and mining solid waste images, such as single feature extraction, difficult fusion, missing high-order features, weak generalization ability and low computational efficiency, an innovative s...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-04-01
|
| Series: | Discover Applied Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s42452-025-06763-2 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Aiming at the problems of traditional methods for multi-dimensional feature extraction of industrial and mining solid waste images, such as single feature extraction, difficult fusion, missing high-order features, weak generalization ability and low computational efficiency, an innovative solution combining Mask R-CNN with Graph Convolutional Networks (GCN) was proposed to achieve automatic, multi-dimensional and efficient feature extraction. The innovation of this method lies in breaking through the bottleneck of multi-feature fusion difficulty in traditional methods by combining instance segmentation and graph structure modeling.The study used Mask R-CNN for instance segmentation of industrial and mining solid waste images, generating masks, bounding boxes, and category information for each instance. Subsequently, primary features such as shape, color, and texture were extracted from the segmentation results, and a graph structure based on the extracted features and relationships between instances was constructed. The graph structure was input into GCN for high-order feature extraction, where the neighbor information of nodes was aggregated through multi-layer graph convolution to update node features, ultimately fusing the high-order features and primary features output by GCN to obtain multidimensional features for classification, detection, and segmentation tasks, thereby improving the accuracy and efficiency of image analysis. The experimental results showed that this method performed well in both processing speed and accuracy, with an average processing time of 108.9 ms, an average accuracy of 96.33%, and an average F1 score of 95.87%. Its overall performance was superior to other mainstream models. This research not only solves the shortcomings of traditional methods in feature extraction and fusion, but also significantly improves computational efficiency and provides new ideas for efficient and automated processing of industrial and mining solid waste images. |
|---|---|
| ISSN: | 3004-9261 |