Contrast Enhancement of UAV-Based Maize Plant Images for Automatic Detection of Fall Armyworm

Automatic plant disease detection involves determining biotic injury caused by pathogens in plants with no direct human contact in the field. One method of automatically detecting plant disease is to deploy unmanned aerial vehicles (UAV) programmed with machine learning algorithms. However, trained machine learning algorithms provide accurate results on when the new testing sample set of plants has the same contrast with the training set. This cannot be guaranteed due to numerous factors. In this paper, we propose a system for automatically detecting infected maize leaves by improving the contrast of maize leaves captured in other farms to correlate with the images used to train the hybrid convolution neural network (HCCN) model. Our experiments show that adjusting the contrast of cropped images from different farms aids in misclassification reduction, particularly when the targeted classes of maize leaves are used as the reference. Specifically, we show that system's performance is improved when the contrast of the healthy and infected images is used.

Automated summary

Automatic plant disease detection can be achieved by deploying unmanned aerial vehicles (UAV) programmed with machine learning algorithms. Improving the contrast of captured images can aid in reducing misclassification.