Self-Supervised Hierarchical Dilated Transformer Network for Hyperspectral Soil Microplastic Identification and...

Area of Science:

• Environmental Science
• Computer Science
• Remote Sensing

Background:

• Microplastics pose significant environmental risks, necessitating effective detection methods.
• Accurate identification of microplastics in soil is crucial for assessing their distribution and ecological impact.
• Current hyperspectral image classification methods often rely on supervised learning, which requires extensive labeled data.

Purpose of the Study:

• To develop an advanced hyperspectral image classification model for microplastic detection in soil.
• To address the challenge of limited labeled training data in soil microplastic identification.
• To enhance feature extraction capabilities for improved microplastic detection accuracy.

Main Methods:

• Proposed the Self-Supervised Hierarchical Dilated Transformer Network (SHDTNet).
• Utilized self-supervised contrastive learning to overcome insufficient training samples.
• Enhanced the feature extraction module within the contrastive learning framework.

Main Results:

• SHDTNet accurately recognizes unique microplastics in soil environments.
• The model demonstrated superior performance compared to existing soil microplastic detection methods.
• Experiments showed no errors or missed detections in self-constructed datasets.

Conclusions:

• SHDTNet offers a robust solution for microplastic detection in soil using hyperspectral imaging.
• Self-supervised learning effectively mitigates the need for large labeled datasets in this domain.
• The proposed model significantly advances the capabilities for environmental microplastic monitoring.