Skip to content

[course project for 2021 Spring<deep learning>.]Identifying multiple diseases on plant leaves by two kinds of modeling. One assuming there are correlation between different diseases one not.

License

Notifications You must be signed in to change notification settings

llxblhyvia/Plant-Pathology

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Plant Pathology

Course project of "Deep Learning", 2021 Spring.

Abstract

The task is from the Kaggle competition "plant-pathology". It aims to identify multiple diseases on plant leaves.

According to our EDA and the research in botany, we found that it seems that there are no correlation between diverse diseases on leaves but we still want to build our model to examine the theory. So we choose two kinds of assumptions to build our CV model. One assuming there are correlation between different diseases while one not. The former is based solely on CNN--different versions of ResNet and DenseNet 169. The latter is based on CNN-RNN architecture.

  • Built several CNN models(ResNet50, ResNet101,DenseNet169) based on the assumption that labels were independent; achieved the highest accuracy of 0.9046 with DenseNet169.
  • Proposed another assumption that labels were correlated and applied LSTM and CNN to process texts and images respectively; utilized multi-label soft margin loss as the loss function and achieved an accuracy of 0.8013.

File Introduction

EDA is the Exploratory data analysis of training data including distribution visualization and corralation analysis.

CNN is the assuming-independent model including data processing, train data, and the three models.

CNN+RNN is the assuming-dependent model including standard CV python files: data processing, train, test, model etc.

Conclusion

According to the model performance that the no-correlation modeling outperforms correlation modeling a lot, proving the theory and our EDA results.

About

[course project for 2021 Spring<deep learning>.]Identifying multiple diseases on plant leaves by two kinds of modeling. One assuming there are correlation between different diseases one not.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published