YoY Challenge 2022: Yum or Yuck Butterfly Classification 2022

Main repository for a Kaggle machine learning competition, concerning the creation of a classification model for various butterfly spiecies. These include Black Swallowtail (Papilio polyxenes), Monarc (Danaus plexippus), Pipevine Swallowtail (Battus philenor), Spicebush Swallowtail (Papilio troilus), Eastern Tiger Swallowtail (Papilio glaucus) and Viceroy (Limenitis archippus).

Table of Contents

1. About The Project
   • Network architecture
   • Confusion matrix (binary)
   • Confusion matrix (complete)
   • Results
   • Complete project report (italian language)
2. Technologies
3. License
4. Acknowledgments

About The Project

This neural network-based computer vision project is aimed at classifying six different species of butterflies spread in various parts of the planet. The system was developet to participate in a Kaggle challenge, known as Yum or Yuck Butterfly Mimics 2022. The ultimate goal of the latter is the development of reliable systems for automatic butterfly recognition in order to order to be able to conduct detailed studies on the distribution and populations of these insect species, some of which are considered endangered. The results obtained with some state-of-the-art convolutional systems are encouraging despite the limitations of the available hardware, achieving about 94%/95% accuracy on the test set.

Network architecture

YoYNet

ButterflyNet

Confusion matrix of the best model (binary)

Confusion matrix of the best model (complete)

Results

#	Size/Version	Freeze pre-trained weights	Dense dimension	Bias regularizer
1	B0	Sì	$\left[512, 128, 6\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$
2	B0	No	$\left[512, 256, 6\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$
3	B2	No	$\left[512, 6\right]$	$\left[L_1: 10^{-6}, -\right]$
4	B1	No	$\left[512, 256, 6\right]$	$\left[L_2: 10^{-5}, L_2: 10^{-5}, -\right]$
5	B1	No	$\left[512, 6\right]$	$\left[ -, -\right]$
6	B1	No	$\left[512, 512, 6\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$
7	B1	No	$\boldsymbol{\left[512, 256, 6\right]}$	$\boldsymbol{\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]}$
8	B1	No	$\left[512, 512, 6\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$
9	VGG19	No	$\left[512, 6\right]$	$\left[-, -\right]$
10	VGG19	No	$\left[512, 256, 6\right]$	$\left[ L_1: 10^{-6}, L_1: 10^{-6}, -\right]$
11	VGG16	No	$\left[512, 256, 6\right]$	$\left[L_1: 10^{-5}, L_1: 10^{-5}, -\right]$
12	VGG16	No	$\left[512, 6\right]$	$\left[L_2: 10^{-5},-\right]$

#	Weight regularizer	Activation regularizer
1	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$	$\left[-, L_1: 10^{-6}, L_1: 10^{-6}\right]$
2	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$	$\left[-, L_1: 10^{-6}, L_1: 10^{-6}\right]$
3	$\left[L_1: 10^{-6}, -\right]$	$\left[-, L_1: 10^{-6}\right]$
4	$\left[L_2: 10^{-5}, L_2: 10^{-5}, -\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, L_1: 10^{-6}\right]$
5	$\left[L_1: 10^{-6}, -\right]$	$\left[-, -\right]$
6	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, L_1: 10^{-6}\right]$
7	$\boldsymbol{\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]}$	$\boldsymbol{\left[L_1: 10^{-6}, L_1: 10^{-6}, L_1: 10^{-6}\right]}$
8	$\left[L_1: 10^{-5}, L_1: 10^{-6}, -\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6}, L_1: 10^{-6}\right]$
9	$\left[-, -\right]$	$\left[-, -\right]$
10	$\left[L_1: 10^{-6}, L_1: 10^{-6}, -\right]$	$\left[L_1: 10^{-6}, L_1: 10^{-6},
11	$\left[L_1: 10^{-5}, L_1: 10^{-5}, -\right]$	$\left[L_1: 10^{-5}, L_1: 10^{-5}, L_1: 10^{-5}\right]$
12	$\left[L_2: 10^{-5}, -\right]$	$\left[-, L_1: 10^{-6}\right]$

#	Optimizer	$\boldsymbol{\eta}$
1	Adadelta	1
2	Adadelta	1
3	Adadelta	1
4	Adam	$3 \cdot 10^{-7}$
5	Adadelta	1
6	Adadelta	1
7	Adadelta	1
8	Adadelta, Adam	1, $3 \cdot 10^{-7}$
9	Adam	$3 \cdot 10^{-7}$
10	Adam	$3 \cdot 10^{-7}$
11	Adam	$3 \cdot 10^{-7}$
12	Adam	$3 \cdot 10^{-6}$

#	Precision	Recall	Accuracy	F1-score
1	88.30%	88.51%	89.78%	88.11%
2	91.18%	91.28%	91.25%	91.24%
3	92.58%	92.25%	92.70%	92.40%
4	86.74%	86.55%	86.55%	86.28%
5	92.10%	91.23%	91.23%	91.21%
6	94.88%	93.97%	94.15%	94.12%
7	96.64%	96.49%	96.49%	96.48%
8	94.30%	94.15%	94.74%	94.68%
9	83.58%	83.34%	83.63%	82.27%
10	90.69%	90.97%	90.64%	90.47%
11	88.82%	88.08%	88.89%	87.89%
12	88.30%	88.66%	88.89%	88.26%

Complete project report (italian language)

Link

(back to top)

Technologies

Deep learning models

• 
• 

Preprocessing and data management

• 
• 
• 
• 

Data visualization

• 

License

Distributed under the Apache License 2.0 License. See LICENSE for more information.

(back to top)

Acknowledgments

• Link to the Kaggle competition
• Best-README template GitHub repository

(back to top)