An improved algorithm for salient object detection of microscope based on U²-Net.

Area of Science:

• Medical Imaging
• Computer Vision
• Deep Learning

Background:

• Manual microscopy is inefficient and time-consuming.
• Accurate medical image analysis is crucial for diagnosis and research.
• Current methods lack efficiency and accuracy in salient object detection.

Purpose of the Study:

• To develop an improved deep learning algorithm for salient object detection in microscopy images.
• To enhance the efficiency and accuracy of medical image capture and analysis.
• To reduce the computational burden of quantitative analysis in microscopy.

Main Methods:

• An improved salient object detection algorithm based on U^2-Net was developed.
• Incorporated Convolutional Block Attention Module (CBAM) for enhanced feature extraction.
• Optimized network complexity with Simple Pyramid Pooling Module (SPPM) and Ghost convolution for model lightweighting.
• Applied data augmentation to improve robustness and generalization.

Main Results:

• The improved model size was reduced by 56.85% (72.5 MB vs. 168.0 MB).
• Prediction accuracy increased from 92.24% to 97.13%.
• Demonstrated significant improvements in model size and prediction accuracy.

Conclusions:

• The proposed algorithm offers an efficient and accurate solution for salient object detection in microscopy.
• The lightweight and accurate model facilitates subsequent image processing and quantitative analysis.
• Deep learning advancements show promise for improving microscopy-based medical image analysis.