A Convolutional Neural Network for Automatic Identification and Classification of Fall Army Worm Moth

To combat the problem caused by the Fall Army Worm in the country there is a need to come up with robust early warning and monitoring systems as the current manual system is labor intensive and time consuming. The automation of the identification and classification of the insect is one of the novel methods that can be undertaken. Therefore this paper presents the results of training a Convolutional Neural Network model using Google’s Tensorflow Deep Learning Framework for the identification and classification of the Fall Army worm moth. Due to lack of enough training dataset and good computing power, we used transfer learning, which is the process of reusing a model trained on one task as a starting point for a model on a second task. Googles pre-trained InceptionV3 model was used as the underlying model. Data was collected from four sources namely the field, Lab setup, by crawling the internet and using Data Augmentation. We Present results of the best three trials in terms of training accuracy after several attempts to get the best metrics in terms of learning rate and training steps. The best model gave a prediction average accuracy of 82% and a 32% average prediction accuracy on false positives. The results shows that it is possible to automate the identification and classification of the Fall Army worm Moth using Convolutional Neural Networks.

@article{Chulu2019,
  title     = {A Convolutional Neural Network for Automatic Identification and Classification of Fall Army Worm Moth},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {10},
  number    = {7},
  year      = {2019},
  publisher = {The Science and Information Organization},
  author    = {Francis Chulu and Jackson Phiri and Phillip O.Y. Nkunika and Mayumbo Nyirenda and Monica M.Kabemba and Philemon H.Sohati},
  doi       = {10.14569/IJACSA.2019.0100717},
  url       = {https://doi.org/10.14569/IJACSA.2019.0100717}
}