An embedded, machine learning-enabled platform for in-field screening of plant disease and pest damage.

Abstract

The fall armyworm pest caught farmers and agricultural organizations by surprise when it hit most major maize farming regions in Uganda. Currently, both large-scale and small-scale farmers rely on manual observation of the maize crop for detection of the Fall Armyworm. This comes weeks after the pest has fully matured and began causing damage to crops. The late detection of the Fall Armyworm in turn results to delays in administering effective pest control measures which forces farmers to incur high costs in administering appropriate control measures. With the ineffectiveness of late manual observations, there is need for an early technology-based solution that will allow farmers to prepare in advance for possible Fall Armyworm infestations. The project aimed at developing a classification model that detects fall army worm damages in maize using transfer learning. The dataset used for the model development was obtained from the Marconi society machine learning laboratory, it had 2,818 curated images belonging to two classes (fallarmyworm and healthy). Mobile Net weights were used to train the model and the performance metrics used were accuracy, recall, precision and F1 score. The model was then deployed on the AI enabled platform that was developed using Qt frameworks on raspberry pi 4. The system can be used for early detection of diseases and pest damages using machine learning models for various plants.