Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AquaFusionNet: Lightweight VisionSensor Fusion Framework for Real-Time Pathogen Detection and Water Quality Anomaly Prediction on Edge Devices

Sepyan Purnama Kristanto, Lutfi Hakim, Hermansyah

TL;DR

This work tackles real-time pathogen detection and water-quality anomaly prediction in small-scale drinking-water systems by unifying microscopic imagery with short-horizon sensor data on edge devices. It introduces AquaFusionNet, a hardware-aware fusion framework featuring a pruned vision branch, a compact AquaTemp-Net temporal branch, and gated cross-attention to fuse modalities, trained end-to-end and quantized for low-power deployments. The authors present AquaMicro12K, a large annotated microscopic dataset, and validate the approach through a six-month field deployment across seven Indonesian facilities, achieving high detection (mAP@0.5) and anomaly-prediction performance with low false-alarm rates. They demonstrate significant benefits of cross-modal coupling over unimodal baselines, emphasize hardware–software co-design, and openly share data, code, and hardware designs to enable replication in decentralized water-safety infrastructures.

Abstract

Evidence from many low and middle income regions shows that microbial contamination in small scale drinking water systems often fluctuates rapidly, yet existing monitoring tools capture only fragments of this behaviour. Microscopic imaging provides organism level visibility, whereas physicochemical sensors reveal shortterm changes in water chemistry; in practice, operators must interpret these streams separately, making realtime decision-making unreliable. This study introduces AquaFusionNet, a lightweight cross-modal framework that unifies both information sources inside a single edge deployable model. Unlike prior work that treats microscopic detection and water quality prediction as independent tasks, AquaFusionNet learns the statistical dependencies between microbial appearance and concurrent sensor dynamics through a gated crossattention mechanism designed specifically for lowpower hardware. The framework is trained on AquaMicro12K, a new dataset comprising 12,846 annotated 1000 micrographs curated for drinking water contexts, an area where publicly accessible microscopic datasets are scarce. Deployed for six months across seven facilities in East Java, Indonesia, the system processed 1.84 million frames and consistently detected contamination events with 94.8% mAP@0.5 and 96.3% anomaly prediction accuracy, while operating at 4.8 W on a Jetson Nano. Comparative experiments against representative lightweight detectors show that AquaFusionNet provides higher accuracy at comparable or lower power, and field results indicate that cross-modal coupling reduces common failure modes of unimodal detectors, particularly under fouling, turbidity spikes, and inconsistent illumination. All models, data, and hardware designs are released openly to facilitate replication and adaptation in decentralized water safety infrastructures.

AquaFusionNet: Lightweight VisionSensor Fusion Framework for Real-Time Pathogen Detection and Water Quality Anomaly Prediction on Edge Devices

TL;DR

This work tackles real-time pathogen detection and water-quality anomaly prediction in small-scale drinking-water systems by unifying microscopic imagery with short-horizon sensor data on edge devices. It introduces AquaFusionNet, a hardware-aware fusion framework featuring a pruned vision branch, a compact AquaTemp-Net temporal branch, and gated cross-attention to fuse modalities, trained end-to-end and quantized for low-power deployments. The authors present AquaMicro12K, a large annotated microscopic dataset, and validate the approach through a six-month field deployment across seven Indonesian facilities, achieving high detection (mAP@0.5) and anomaly-prediction performance with low false-alarm rates. They demonstrate significant benefits of cross-modal coupling over unimodal baselines, emphasize hardware–software co-design, and openly share data, code, and hardware designs to enable replication in decentralized water-safety infrastructures.

Abstract

Evidence from many low and middle income regions shows that microbial contamination in small scale drinking water systems often fluctuates rapidly, yet existing monitoring tools capture only fragments of this behaviour. Microscopic imaging provides organism level visibility, whereas physicochemical sensors reveal shortterm changes in water chemistry; in practice, operators must interpret these streams separately, making realtime decision-making unreliable. This study introduces AquaFusionNet, a lightweight cross-modal framework that unifies both information sources inside a single edge deployable model. Unlike prior work that treats microscopic detection and water quality prediction as independent tasks, AquaFusionNet learns the statistical dependencies between microbial appearance and concurrent sensor dynamics through a gated crossattention mechanism designed specifically for lowpower hardware. The framework is trained on AquaMicro12K, a new dataset comprising 12,846 annotated 1000 micrographs curated for drinking water contexts, an area where publicly accessible microscopic datasets are scarce. Deployed for six months across seven facilities in East Java, Indonesia, the system processed 1.84 million frames and consistently detected contamination events with 94.8% mAP@0.5 and 96.3% anomaly prediction accuracy, while operating at 4.8 W on a Jetson Nano. Comparative experiments against representative lightweight detectors show that AquaFusionNet provides higher accuracy at comparable or lower power, and field results indicate that cross-modal coupling reduces common failure modes of unimodal detectors, particularly under fouling, turbidity spikes, and inconsistent illumination. All models, data, and hardware designs are released openly to facilitate replication and adaptation in decentralized water safety infrastructures.

Paper Structure

This paper contains 30 sections, 5 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem Statement and Motivation
  3. Limitations of Existing Approaches
  4. Contributions
  5. Related Work
  6. IoT and Edge-Based Water Quality Monitoring
  7. Deep Learning for Water Quality Prediction
  8. Microscopic Pathogen and Algae Detection
  9. Multimodal Fusion and Lightweight Edge Models
  10. Materials and Methods
  11. Field Deployment
  12. AquaMicro12K Dataset
  13. AquaFusionNet Architecture
  14. Vision Branch
  15. Temporal Branch: AquaTemp-Net
  16. ...and 15 more sections

Figures (3)

  • Figure 1: Schematic of an AquaFusionNet deployment at a drinking-water depot, showing the water source, pump, sensor manifold, microscopic imaging chamber, edge compute node (ESP32-S3 + Jetson Nano), and solar-powered supply. Water flows through the manifold and microscope chamber before discharge, while sensor and image streams are processed locally on the edge node.
  • Figure 2: AquaFusionNet architecture. The vision branch uses a pruned SSD--MobileNetV3-Small backbone to detect microorganisms and particulate contaminants in microscopic frames. AquaTemp-Net processes multivariate physicochemical time series (pH, turbidity, TDS, temperature, DO, ORP) into a compact temporal descriptor. A gated cross-attention module fuses visual feature $f_v$ and temporal feature $f_t$ to produce a fused representation $f_{\text{fused}}$ used for anomaly prediction, while the SSD head outputs bounding boxes and class scores.
  • Figure 3: Field performance (January--June 2025). Top: daily number of AquaFusionNet alerts overlaid with laboratory-confirmed contamination events. Bottom: false positive and false negative rates by site, illustrating variability across depots and river intakes.