LBMS-SAM: Segment anything model guided SEM image segmentation for lithium battery materials.

Area of Science:

• Materials Science
• Artificial Intelligence
• Image Analysis

Background:

• Quality inspection of lithium battery materials relies on manual analysis of particle sizes in SEM images.
• Manual annotation is time-consuming, labor-intensive, and susceptible to subjective errors.
• Automating this process is crucial for improving efficiency and accuracy in material quality control.

Purpose of the Study:

• To develop an automated artificial intelligence solution for lithium battery material quality inspection.
• To address the limitations of manual annotation in analyzing particle sizes from SEM images.
• To introduce a novel deep learning model for SEM image segmentation tasks.

Main Methods:

• A new dataset, the LBMS dataset, was created specifically for lithium battery material SEM image segmentation (LBMS).
• A specialized model, LBMS-SAM, was proposed, incorporating a Gabor and Sobel edge feature extraction module (GSEFE).
• A multi-layer denoised features fusion module (MDFF) using wavelet transform was designed to enhance feature extraction and reduce noise.

Main Results:

• The proposed LBMS-SAM model demonstrated superior performance on the LBMS dataset.
• LBMS-SAM outperformed existing state-of-the-art (SOTA) methods across all evaluation metrics.
• The model achieved accurate extraction of edge information and efficient fusion of global contextual features with minimal added parameters.

Conclusions:

• The developed LBMS-SAM model offers an effective and automated solution for lithium battery material quality inspection.
• The AI-driven approach significantly enhances accuracy and efficiency compared to traditional manual methods.
• This work paves the way for advanced automated quality control in battery material manufacturing.