Localizing plant leaves using maximum anchor boxes in region proposal convolutional neural networks

Abstract

As the population of the earth grows, the demand for food grows proportionally. Early and cost-effective detection of plant diseases can result in less food loss throughout the world. The current methods for image-based plant disease detection tend to fail in field conditions. Our method uses region proposal networks to localize diseased leaves for detection. We discard no prior anchor boxes, which increases the average recall of the network, resulting in better localization.