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

## TL;DR

WPF-Mamba is a new multispectral detection framework that improves accuracy in identifying small, visually similar microorganisms using wavelet-based fusion and progressive feature refinement.

## Contribution

The novel Progressive Visual State Space Block and Wavelet-based Multispectral Fusion module enhance feature alignment and spectral consistency for microorganism detection.

## Key findings

- WPF-Mamba improves mAP@50 by 2.9% on the EMDS-7-MS dataset compared to baseline models.
- Wavelet-based fusion mitigates cross-band inconsistencies and enhances texture and spectral features.
- Progressive feature refinement improves subtle feature capture for small microorganisms.

## Abstract

Accurate detection of environmental microorganisms is key to ecological monitoring and public health risk assessment. Multispectral imaging yields rich biochemical and structural cues, yet its practical use is hampered by inter-band spectral heterogeneity and the small, visually similar traits of microorganisms objects. These issues impair cross-band feature alignment and discriminability, thus limiting the performance of existing detection frameworks.

To address these challenges, we propose a multispectral framework for fine-grained microorganisms detection named WPF-Mamba (Wavelet-Progressive Fusion Mamba). We design a novel Progressive Visual State Space Block (P-VSS Block). Built on the conventional VSS block, it integrates a Progressive Multi-Scale Feature Fusion (PMFF) unit to optimize hierarchical representations via stepwise context and semantic enhancement, improving subtle feature capture. WPF-Mamba further incorporates a Wavelet-based Multispectral Fusion (WMF) module, which fuses complementary spectral information through multi-scale wavelet decomposition and frequency-domain alignment, mitigating cross-band inconsistencies and enhancing microorganisms texture and spectral feature representation.

Based on the EMDS-7 dataset, we extended the sample set by constructing high-quality infrared samples with generative adversarial networks and generative large language models, thus forming the extended multispectral microorganisms detection dataset EMDS-7-MS. Evaluation results on the EMDS-7-MS dataset demonstrate that our method further improves the mAP@50 by 2.9% compared with the baseline model, which verifies the effectiveness of our proposed method in the task of multispectral microorganisms detection.

By addressing spectral misalignment and small-object representation limitations, WPF-Mamba offers a robust, generalizable approach for multispectral microorganisms detection. Specifically, its wavelet-based fusion and progressive feature refinement strategy presents a practical paradigm for multispectral fine-grained microorganisms analysis, which in turn contributes to the development of reliable, scalable environmental monitoring systems.