Skip to content

Repository files navigation

Hi 👋, I'm Md Suzauddola

Researcher | Deep Learning Engineer | Computer Vision for Agriculture

Research Focus PyTorch YOLO

🌟 Featured Work

Advanced Channel-Enhanced Multi-Scale Pest Detection Network

Published in Expert Systems with Applications (2025)

CMPestNet Overview

Key Achievements

Dataset mAP / Acc Improvement FPS
Jute17 85.6% mAP +9.2% 350+
Pest24 78.22% mAP +8.02% 370+
IP102 78.15% Acc -- --

⭐ Star CMPestNet📄 Read the Paper⬇️ Download Models


🚀 About Me

I am a researcher focused on developing efficient deep learning solutions for real-world agricultural challenges. My work emphasizes crop pest detection in complex field environments, addressing small objects, camouflage, and multi-scale variability.

Research Interests:

  • Object Detection & Computer Vision
  • Efficient Neural Architectures (YOLO-based)
  • Agricultural AI & Precision Farming
  • Multi-scale Feature Extraction & Channel Attention

📌 Featured Project: CMPestNet

"Advanced deep learning model for crop-specific and cross-crop pest identification"

✨ Core Innovations

  • CE-GELAN (Channel-Enhanced Generalized Efficient Layer Aggregation Network)
  • GMSFE-ELAN (Generalized Multi-Scale Feature Extraction with 1×1, 3×3, 5×5, 7×7 kernels)
  • Custom Re-parameterization for real-time inference (350+ FPS)
  • New Jute17 Dataset (12,916 images, 17 pest classes)

📊 Performance Highlights

  • Jute17: 85.6% mAP (9.2% better than baseline)
  • Pest24: 78.22% mAP (excellent on tiny pests)
  • IP102: 78.15% Accuracy (outperforms multiple SOTA models)
  • Strong generalization across crop-specific and cross-crop scenarios

🛠️ Tech Stack

  • Framework: PyTorch 2.1+
  • Architecture: Enhanced YOLO-style with custom modules
  • Tools: OpenCV, Albumentations, Grad-CAM, etc.

📁 Repository Structure (CMPestNet)

CMPestNet/
├── README.md                  # ← You are here
├── LICENSE
├── requirements.txt
├── setup.py
├── assets/                    # Visuals & Results
│   ├── teaser.jpg
│   ├── architecture.png
│   ├── gradcam_comparison.png
│   └── performance_charts/
├── models/
│   ├── cmpestnet.py
│   ├── ce_gelan.py
│   ├── gmsfe_elan.py
│   └── reparam.py
├── utils/
│   ├── augmentations.py
│   ├── loss.py
│   └── metrics.py
├── scripts/
│   ├── train.py
│   ├── infer.py
│   └── export.py
├── notebooks/
│   ├── training_demo.ipynb
│   └── inference_demo.ipynb
├── config/
│   └── hyperparameters.yaml
├── data/                      # Download scripts
└── weights/                   # Pre-trained models (Git LFS)


## 🚀 About Me

# Clone the repo
git clone https://github.com/yourusername/CMPestNet.git
cd CMPestNet

# Install dependencies
pip install -r requirements.txt

# Inference example
python scripts/infer.py --weights weights/cmpestnet_jute17.pt --source test_images/ --save-txt

# Clone the repo
git clone https://github.com/yourusername/CMPestNet.git
cd CMPestNet

# Install dependencies
pip install -r requirements.txt

# Inference example
python scripts/infer.py --weights weights/cmpestnet_jute17.pt --source test_images/ --save-txt

📄 Citation

@article{suzauddola2025cmpestnet,
  title   = {Advanced deep learning model for crop-specific and cross-crop pest identification},
  author  = {Md Suzauddola and Defu Zhang and Adnan Zeb and Junde Chen and others},
  journal = {Expert Systems with Applications},
  year    = {2025},
  doi     = {10.1016/j.eswa.2025.126896}
}

🌐 Connect With Me

GitHub: @suzauddola
Google Scholar / ResearchGate / LinkedIn: https://sites.google.com/view/mdsuzauddola/home

About

Advanced Channel-Enhanced Multi-Scale Pest Detection Network (CMPestNet) for crop-specific and cross-crop pest identification. Outperforms SOTA on Jute17, Pest24, and IP102 datasets.

Topics

Resources

Stars

Watchers

Forks

Packages

Contributors

Languages