A Robust Pest Identification System using Morphological Analysis in Neural Networks

Abstract

Timely  detection  of  pests  play  a  major  role  in  agriculture.   There  exist  many  pest  identification  systems,  but  almost  all  of  them  suffer  from  the  misclassification  due  to  lighting, background clutter, heterogeneous capturing devices as well as the pest being partially visible or in the different orientation.  This misclassification may cause tremendous yield loss. To  mitigate  this  situation,  we  proposed  an  architecture  to  provide  high  classification  accuracy under the aforementioned conditions using morphology and skeletonization along with neural networks as classifiers. We have considered the crop rice as a use case as it is the staple food grain of almost the entire population of India. The amount of pesticides used is highest in rice as compared to all other food grains. This paper offers a robust technique to identify the pests in rice crops. The performance of the proposed architecture is tested with an image dataset, and the experimental results reveal that our proposed approach provides better classification accuracy than the existing pest detection approaches in the literature. Furthermore, the experimental results also provide the performance comparison among the popular classifiers.