Coarse Clustering and Classification of Images with CNN Features for Participatory Sensing in Agriculture

Prakruti Bhatt, Sanat Sarangi, Srinivasu Pappula

Abstract

A solution is proposed to perform unsupervised image classification and tagging by leveraging the high level features extracted from a pre-trained Convolutional Neural Network (CNN). It is validated over images collected through a mobile application used by farmers to report image-based events like pest and disease incidents, and application of agri-inputs towards self-certification of farm operations. These images need to be classified into their respective event classes in order to help farmers tag images properly and support the experts to issue appropriate advisories. Using the features extracted from CNN trained on ImageNet database, images are coarsely clustered into classes for efficient image tagging. We evaluate the performance of different clustering methods over the feature vectors of images extracted from global average pooling layer of state-of-the-art deep CNN models. The clustered images represent a broad category which is further divided into classes. CNN features of the tea leaves category of images were used to train the SVM classifier with which we achieve 93.75% classification accuracy in automated state diagnosis of tea leaves captured in uncontrolled conditions. This method creates a model to auto-tag images at the source and can be deployed at scale through mobile applications.

Download


Paper Citation


in Harvard Style

Bhatt P., Sarangi S. and Pappula S. (2018). Coarse Clustering and Classification of Images with CNN Features for Participatory Sensing in Agriculture.In Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM, ISBN 978-989-758-276-9, pages 488-495. DOI: 10.5220/0006648504880495


in Bibtex Style

@conference{icpram18,
author={Prakruti Bhatt and Sanat Sarangi and Srinivasu Pappula},
title={Coarse Clustering and Classification of Images with CNN Features for Participatory Sensing in Agriculture},
booktitle={Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,},
year={2018},
pages={488-495},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0006648504880495},
isbn={978-989-758-276-9},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 7th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,
TI - Coarse Clustering and Classification of Images with CNN Features for Participatory Sensing in Agriculture
SN - 978-989-758-276-9
AU - Bhatt P.
AU - Sarangi S.
AU - Pappula S.
PY - 2018
SP - 488
EP - 495
DO - 10.5220/0006648504880495