RhizoWhisperer: RHIZO-NET Root Health & Edaphic Topology Optimization Network

RhizoWhisperer (RHIZO-NET) is an end-to-end deep learning framework, topological graph phenotyping engine, and edaphic climate-resilience platform for precision agriculture. It integrates custom computer vision models, soft skeletonization topology extraction, multi-modal PyG GNN fusion, ISRIC SoilGrids chemical profiling, and TNAU agronomic recommendation engines.

---

📑 Table of Contents

• Key Features & Scientific Innovations
• System Architecture
• Repository Layout
• Prerequisites & Installation
• Quickstart & Pipeline Execution
• Benchmark Results
• Model Architectures Repository
• License & Citation

---

🌟 Key Features & Scientific Innovations

1. Custom Neural Architecture Suite

• RhizoAttentionNet: Features Oriented Topological Attention Modules (OTAM) + Multi-Scale Receptive Field Pyramids (MSRFP). Achieves 97.9% IoU and 0.0412 Loss.
• DualStreamRootNet: Dual-encoder fusing spatial RGB features with tubular vesselness features extracted via multi-scale Frangi Hessian filtering.
• RhizoHybridTransformer: Ultra-lightweight Swin Shifted-Window Transformer with Root Query Tokens (RQT), achieving 95.8% IoU with only 79.7K parameters (1.8 ms inference latency).
• RhizoGraphFormer: Graph Transformer utilizing Laplacian Positional Encoding (LPE) to capture global topological connectivity across root junction nodes.

2. Novel Physics-Informed & Topology-Preserving Loss Suite

• clDiceLoss: Centerline Dice loss enforcing topological connectivity across fine root structures.
• PIET-Loss (Physics-Informed Edaphic Transport): Enforces physical mass-conservation of water/nutrient flux along root centerlines ($\nabla \cdot \mathbf{J} = 0$).
• TopologyAwareLoss: Multi-term composite loss integrating BCE, Dice, Focal, clDice, and PIET-Loss penalties.

3. Real-World Climate & Agronomic Modules

• Generative Root Skeleton Reconstruction (GRSR): Morphological gap repair algorithm connecting disconnected segments caused by soil particle occlusion.
• CARRS (Climate-Adaptive Root Resiliency Simulator): Models root hydraulic conductivity ($K_{rh}$) and drought vulnerability under RCP 4.5 and RCP 8.5 warming scenarios.
• RCS-Flux (Rhizosphere Carbon Sequestration Predictor): Estimates annual root carbon input ($C_{root}$) and carbon credit financial yield ($35.20/ha/year).
• TNAU Agronomic Engine: Rule-based prescription system providing custom fertilizer schedules and lockout remediation for Sorghum, Tomato, Turmeric, Groundnut, and African Marigold.

---

🏗️ System Architecture

flowchart TD
    subgraph DataIngestion ["Stage 1: Multi-Dataset Ingestion (106.9K Images)"]
        D1[RootNav 2.0] & D2[PRMI Collection] & D3[DeepRootLab] & D4[SeminalRootAngle] & D5[Chicory] & D6[Grassland]
    end

    subgraph CoreSegmentation ["Stage 2-3: Deep Neural Segmentation"]
        D1 & D2 & D3 & D4 & D5 & D6 --> M[RhizoAttentionNet / RhizoUNet]
        M -->|Low Confidence < 0.50| SAM[MobileSAM Fallback Adapter]
        M -->|Predicted Mask| GRSR[GRSR Gap Reconstruction Module]
        SAM --> GRSR
    end

    subgraph TopologyExtract ["Stage 4-5: skan Phenotyping & Graph Transformer"]
        GRSR --> SKEL[skan Skeletonization]
        SKEL --> FEAT[Morphometric Extraction: Length, Tortuosity, Sholl, Angle]
        SKEL --> RGF[RhizoGraphFormer Laplacian Positional Encoding]
    end

    subgraph MultiModalFusion ["Stage 6-8: PyG GNN & SoilGrids Fusion"]
        SG[ISRIC SoilGrids 0-200cm Depth Chemistry] --> FUSION[PyG 2.0 GNN + RhizoFusionNet]
        FEAT & RGF --> FUSION
        FUSION --> DIAG[Nutrient Deficiency Diagnosis]
    end

    subgraph AgronomicClimate ["Stage 9-15: Prescriptions & Climate Simulation"]
        DIAG --> TNAU[TNAU Agronomic Engine]
        TNAU --> PRES[Multi-Crop NPK Prescriptions & Lockout Protocols]
        FEAT --> CARRS[CARRS Climate Drought Simulator]
        FEAT --> RCS[RCS Carbon Sequestration & ROI Calculator]
    end

Loading

---

📂 Repository Layout

RhizoWhisperer/
├── README.md                           # Master Project Documentation
├── CONTRIBUTING.md                     # Contribution Guidelines
├── LICENSE                             # Apache License 2.0
├── requirements.txt                    # Project Dependencies
├── kaggle_version_execution_flowcharts.md # Mermaid Flowcharts (v1 to v15)
│
├── src/                                # Core Source Code
│   ├── unet/                           # Neural Architectures & Loss Suite
│   │   ├── model.py                    # RhizoUNet
│   │   ├── rhizo_attention_net.py      # RhizoAttentionNet (OTAM + MSRFP)
│   │   ├── dual_stream_root_net.py     # DualStreamRootNet (Hessian Dual)
│   │   ├── rhizo_hybrid_transformer.py # RhizoHybridTransformer (Swin + RQT)
│   │   └── losses.py                   # TopologyAwareLoss & PIET-Loss
│   ├── topology/                       # skan Graph Extraction & GRSR
│   │   ├── skeletonize.py              # Medial Axis Skeletonization
│   │   ├── graph_extract.py            # NetworkX Graph Construction
│   │   ├── phenotype_features.py       # Tortuosity & Sholl Analysis
│   │   ├── seminal_angle.py            # Seminal Opening Angle
│   │   └── reconstruct.py              # GRSR Gap Repair Module
│   ├── fusion/                         # Multi-Modal GNN & Graph Transformer
│   │   ├── graph_transformer.py        # RhizoGraphFormer (LPE)
│   │   ├── gnn_encoder.py              # PyG GNN Encoder
│   │   ├── tensor_frame_model.py       # RhizoFusionNet
│   │   └── soil_features.py            # Soil Vector Normalizer
│   ├── agronomic/                      # TNAU Engine & ROI Calculator
│   │   ├── recommendation_engine.py    # Multi-Crop Prescriptions
│   │   ├── soil_rating.py              # Soil Fertility Index
│   │   ├── crop_profiles.py            # Crop Blanket NPK Standards
│   │   └── roi_calculator.py           # Financial ROI Calculator
│   └── climate/                        # Real-World Climate Simulators
│       ├── resiliency_simulator.py     # CARRS Drought Simulator
│       └── carbon_sequestration.py     # RCS Carbon Credit Flux
│
├── architecture/                       # Exported ONNX Binaries & Scripts
│   ├── rhizo_unet.onnx                 # (6.69 MB)
│   ├── rhizo_attention_net.onnx        # (22.55 MB)
│   ├── dual_stream_root_net.onnx       # (18.73 MB)
│   ├── rhizo_hybrid_transformer.onnx   # (1.17 MB)
│   └── export_all_onnx.py              # ONNX Export Script
│
├── notebooks/                          # Interactive Jupyter Notebooks
│   ├── 01_data_preparation.ipynb
│   ├── 02_unet_root_segmentation.ipynb
│   ├── 03_mobilesam_segmentation.ipynb
│   ├── 04_topology_extraction.ipynb
│   ├── 05_multimodal_fusion.ipynb
│   ├── 06_agronomic_engine.ipynb
│   ├── 07_full_pipeline_demo.ipynb
│   └── kaggle_push/
│       └── rhizo_net_all_stages_run.py # 15-Stage Automated Kaggle Runner
│
└── kaggle_outputs/                     # Extracted Logs & 25 PNG Plots
    ├── version_10/                     # 7-Stage Execution Log
    ├── version_12/                     # 12-Stage Execution Log
    ├── version_14/                     # 18 PNG Plot Files
    └── version_15/                     # 25 PNG Plot Files & Full Log

---

⚡ Prerequisites & Installation

Requirements

• Linux or macOS (Ubuntu 20.04+, macOS 12+)
• Python 3.10 or higher
• PyTorch 2.0+ with CUDA or CPU support

# Clone the repository
git clone https://github.com/Runtime-Slayers/RhizoWhisperer.git
cd RhizoWhisperer

# Set up virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

---

🚀 Quickstart & Pipeline Execution

Execute the full 15-stage automated deep learning and phenotyping pipeline:

python3 notebooks/kaggle_push/rhizo_net_all_stages_run.py

All 25 high-resolution PNG plots will be generated and saved to ./output_plots/.

---

📊 Benchmark Results

Model Architecture	Parameters	Loss (Curriculum)	Segmentation IoU	ONNX Latency (CPU)	Primary Target Platform
RhizoUNet	1,746,737	0.0580	94.2%	4.2 ms	Edge Servers / Workstations
DualStreamRootNet	4,885,959	0.0482	96.5%	9.6 ms	Soil Rhizotron Analysis
RhizoHybridTransformer	79,749	0.0451	95.8%	1.8 ms	Mobile & Field Drones
RhizoAttentionNet	5,892,305	0.0412	97.9%	12.8 ms	GPU Cloud Clusters

---

🏛️ Model Architectures Repository

For dedicated PyTorch source code, layer specifications, receptive field calculations, and ONNX Runtime benchmark scripts, visit our dedicated sub-repository:

👉 Runtime-Slayers/RhizoWhisperer-Model-Architectures

---

📜 License & Citation

Distributed under the Apache License 2.0. See LICENSE for details.

Citation

@article{runtime_slayers_rhizowhisperer_2026,
  title={RHIZO-NET: Root Health and Integrated Zonal Optimization Network via Edaphic Topology},
  author={Runtime Slayers Team},
  year={2026},
  publisher={GitHub Repository},
  url={https://github.com/Runtime-Slayers/RhizoWhisperer}
}