Developing an automated fall armyworm identification, early warning and monitoring system using a convolution neural network.

Since its reported presence in 2016, the Fall Armyworm (FAW) has caused major damage to a number of plant species including maize which is a staple food for most African countries. Their presence in Africa pauses a challenge to the food security in many countries contributing to the already existing food problem that the continent has been facing. The outbreak of FAW if left uncontrolled is likely to have a devastating impact on the food supply. Zambia has not been spared by this scourge and currently both commercial and subsistence farmers in Zambia use a manual process to monitor crops for the presence of the fall army worm. Manual monitoring and observation is time consuming, labour intensive and also results in the delay of applying appropriate pest control measures by farmers. There is need to improve on the methods used in the monitoring of the pest if there is to be a proactive and reactive response by stakeholders to FAW.
This study was therefore aimed at improving on the FAW pest monitoring and early warning based on cloud technology using a Convolutional Neural Network model. The objectives were to (1) develop a model using Convolutional Neural Networks for FAW identification, analysis and classification, (2) develop web and mobile applications integrated with GIS technology based on the developed model in the first objective for FAW identification, analysis, processing and visualization.
A Convolutional Neural Network (CNN) was successfully trained to meet objective number (1). Due to lack of enough training dataset and good computing power, a CNN technique called transfer learning was used, by using Googles pre-trained InceptionV3 model as the underlying model. Objective number (2) was addressed by using the developed model in (1) and build a web and a mobile application for data processing, analysis and visualization.
Initial results showed that it is possible to build a Convolutional Neural Network model for automatic identification and classification of the FAW pest based on images acquired from field traps and use the developed model to build web and mobile application for data analysis and real time early warning of the pest occurrence. The developed system provides a real time early warning of the pest occurrence for farmers and stakeholders that never existed before in the country
Key words: Artificial Intelligence, Convolutional Neural Network, FAW, Mobile application, Web, Early warning