Machine-based mosquito taxonomy with a lightweight network-fused efficient dual ConvNet with residual learning and Knowledge Distillation

• Author: Dr. Francis Jesmar P. Montalbo

• Affiliation: Batangas State University

• Email: [email protected]; [email protected]

Personal Webpage

PLEASE CONTANCT ME IF YOU ARE HAVING TROUBLE. I CAN OFFER ASSITANCE

Please do care to cite my works related to this repository

• F. J. P. Montalbo, "Machine-based Mosquito Taxonomy with a Lightweight Network-fused Efficient Dual ConvNet with Residual Learning and Knowledge Distillation," Applied Soft Computing, January, 2023. doi: 10.1016/j.asoc.2022.109913.

• F. J. P. Montalbo, "Automating Mosquito Taxonomy by Compressing and Enhancing a Feature Fused EfficientNet with Knowledge Distillation and a Novel Residual Skip Block," MethodsX, January, 2023. doi: 10.1016/j.mex.2023.102072.

Graphical Abstract

Datasets used:

Mosquito Dataset

Paper to cite:

Classification and Morphological Analysis of Vector Mosquitoes using Deep Convolutional Neural Networks

=========================================================================

❗For a faster method, you may download the already prepared dataset used in the given link below.

CLICK THIS FOR THE PREPARED DATASET USED IN THIS STUDY. Download the data.rar and extract it to the ../dataset

=========================================================================

How to use:

Disclaimer

❗If training the model, the dependencies included a tensorflow-gpu. You may change the tensorflow-gpu to tensorflow if no GPU is to be used. However, the results from the paper were produced using a GPU (RTX 3060 12gb) and may have slight differences

Dependencies included in the requirements.txt:

• jupyter==1.0.0
• keras==2.4.3
• matplotlib==3.4.1
• numpy==1.19.5
• opencv-python==3.4.11.41
• pandas==1.2.4
• Pillow==8.2.0
• scikit-learn==0.24.1
• scikit-image==0.18.1
• scikit-plot==0.3.7
• scipy==1.2.0
• tf-nightly-gpu==2.6.0 (Note: This is optional and can train even with just a CPU or tensorflow non-gpu variant. Nightly is used to compensate the new RTX 3060 card)

=========================================================================

General Instruction:

You may clone using git or download the repository and extract the files manually:

• Once cloned, CD into the folder and enter pip install -r requirements.txt.
• Download the readily trained weights and dataset here ---> Dataset and Trained Weights
• Extract the data.rar in ../dataset and the models.rar in ../models ======================================================================