WPF-Mamba: wavelet-based progressive multispectral fusion mamba for fine-grained microorganism detection

Area of Science:

• Environmental Science
• Microbiology
• Computer Vision

Background:

• Accurate detection of environmental microorganisms is crucial for ecological monitoring and public health.
• Multispectral imaging offers rich data but faces challenges with spectral heterogeneity and small, similar-looking microorganisms.
• Existing detection methods struggle with cross-band feature alignment and discriminability.

Purpose of the Study:

• To propose a robust multispectral framework for fine-grained microorganism detection.
• To overcome limitations in spectral misalignment and small-object representation in existing methods.
• To enhance the accuracy and generalizability of environmental microorganism detection.

Main Methods:

• Developed the Wavelet-Progressive Fusion Mamba (WPF-Mamba) framework.
• Introduced a Progressive Visual State Space Block (P-VSS Block) with Progressive Multi-Scale Feature Fusion (PMFF) for enhanced feature representation.
• Incorporated a Wavelet-based Multispectral Fusion (WMF) module for spectral information fusion and alignment.

Main Results:

• The WPF-Mamba framework demonstrated improved performance on the EMDS-7-MS dataset.
• Achieved a 2.9% increase in mAP@50 compared to the baseline model.
• Validated the effectiveness of the proposed method for multispectral microorganism detection.

Conclusions:

• WPF-Mamba provides a robust and generalizable solution for multispectral microorganism detection.
• The wavelet-based fusion and progressive feature refinement strategy offers a practical paradigm for fine-grained analysis.
• The approach contributes to developing reliable and scalable environmental monitoring systems.