Plant Leaf Disease Recognition Using Fastai Image Classification

Agricultural productivity has a vital role in the Indian economy, but it is seriously hampered by pests and plant diseases. Neural networks have been a major step forward in solving this problem in the past two decades. However, the existing systems in place are computation-heavy and costly to imple...

Full description

Saved in:
Bibliographic Details
Published in2021 5th International Conference on Computing Methodologies and Communication (ICCMC) pp. 1624 - 1630
Main Authors Chakraborty, Aditya, Kumer, Debarun, Deeba, K
Format Conference Proceeding
LanguageEnglish
Published IEEE 08.04.2021
Subjects
Computational modeling
Convolutional neural network (CNN)
Deep learning
densenet-121
fastai
image classification
Image recognition
Neural networks
plantdoc
Plants (biology)
Productivity
Training
transfer learning
Online AccessGet full text

Cover

More Information
Summary:Agricultural productivity has a vital role in the Indian economy, but it is seriously hampered by pests and plant diseases. Neural networks have been a major step forward in solving this problem in the past two decades. However, the existing systems in place are computation-heavy and costly to implement. Such tasks also ideally require a dataset of leaf images that simulates real environment conditions, which is hard to find. The motivation of this paper therefore is to solve all these issues by building a light-weight and cost efficient deep learning architecture with the proposed DenseNet-121 model that classifies leaf images from a dataset called 'PlantDoc' across 28 classes with 1874 training images and 468 validation images. A separate test dataset is held out only for checking model performance on unknown data. Implementation is done using Fastai framework, because of its faster computational power, easy workflow and unique data cleaning functionalities. Overall, the classification accuracy achieved is 92.5%.
DOI:10.1109/ICCMC51019.2021.9418042