Thanks to visit codestin.com
Credit goes to github.com

Skip to content

MosQNet-SA is a novel deep learning model designed for accurate classification of mosquito species. This project focuses on distinguishing between three primary mosquito species: Anopheles, Aedes, and Culex, which are significant vectors for various diseases.

Notifications You must be signed in to change notification settings

masud1901/MosQNet-SA

Repository files navigation

MosQNet-SA: Advanced Mosquito Species Classification

Overview

MosQNet-SA is a novel deep learning model designed for accurate classification of mosquito species. This project focuses on distinguishing between three primary mosquito species: Anopheles, Aedes, and Culex, which are significant vectors for various diseases.

MosQNet-SA Architecture

Dataset

The dataset comprises 3,000 meticulously selected and annotated mosquito images:

  • Anopheles: 1,000 images
  • Aedes: 1,000 images
  • Culex: 1,000 images

Sources:

  1. Mosquito Alert
  2. Mendeley Data
  3. IEEE DataPort
  4. Dryad

The curated dataset used in this study is available on Kaggle: MosQNet-SA Dataset on Kaggle

These are the sample mosquito images MosQNet-SA Architecture

Methodology

Transfer Learning Analysis

We conducted an extensive analysis of various pre-trained models:

  • VGG16, VGG19
  • ResNet50, ResNet101, ResNet152
  • Xception
  • InceptionV3, InceptionResNetV2
  • MobileNet, MobileNetV2
  • DenseNet121, DenseNet169, DenseNet201
  • NASNetMobile
  • EfficientNetB0, EfficientNetB1, EfficientNetB2

MosQNet-SA Architecture

Our proposed model, MosQNet-SA, incorporates:

  • Residual blocks
  • Inception-like blocks
  • MBConv blocks
  • Spatial Attention mechanism

Results

MosQNet-SA outperforms traditional transfer learning approaches:

Model Test Accuracy Params Trainable Params
MosQNet-SA 99.42% 388,349 384,155 (98.92%)

Performance Metrics

Metric Anopheles Aedes Culex
Accuracy 0.987752 0.998721 0.973467
Precision 0.993664 0.990643 0.979466
Recall 0.980208 0.99882 0.985537
F1-score 0.98689 0.994715 0.982492

Data Augmentation

We employed various augmentation techniques:

  • Width and Height Shift (10%)
  • Shear and Zoom Transformations (10%)
  • Horizontal Flipping
  • Nearest Neighbor Filling

Training Strategies

  • Optimizer: Adam
  • Learning Rate: 0.001
  • Epochs: 81

Callbacks

  • EarlyStopping
  • ModelCheckpoint
  • TensorBoard
  • ReduceLROnPlateau
  • CSVLogger
  • LearningRateScheduler

Explainable AI (XAI)

We utilized multiple XAI techniques for model interpretation:

  • Saliency
  • GradientInput
  • GuidedBackprop
  • IntegratedGradients
  • SmoothGrad
  • SquareGrad
  • VarGrad
  • GradCAM
  • Occlusion
  • RISE
  • SobolAttributionMethod
  • LIME
  • KernelShap

XAI Example

Conclusion

MosQNet-SA demonstrates superior performance in mosquito species classification, achieving 99.42% accuracy with a significantly smaller model size compared to traditional transfer learning approaches.

Future Work

  • Expand the dataset to include more mosquito species
  • Explore deployment on edge devices for real-time classification
  • Investigate the model's performance in real-world scenarios

License

This project is licensed under the MIT License - see the LICENSE file for details.

About

MosQNet-SA is a novel deep learning model designed for accurate classification of mosquito species. This project focuses on distinguishing between three primary mosquito species: Anopheles, Aedes, and Culex, which are significant vectors for various diseases.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published