Image set for deep learning: field images of maize annotated with disease symptoms

Abstract

ObjectivesAutomated detection and quantification of plant diseases would enable more rapid gains in plant breeding and faster scouting of farmers’ fields. However, it is difficult for a simple algorithm to distinguish between the target disease and other sources of dead plant tissue in a typical field, especially given the many variations in lighting and orientation. Training a machine learning algorithm to accurately detect a given disease from images taken in the field requires a massive amount of human-generated training data.Data descriptionThis data set contains images of maize (Zea mays L.) leaves taken in three ways: by a hand-held camera, with a camera mounted on a boom, and with a camera mounted on a small unmanned aircraft system (sUAS, commonly known as a drone). Lesions of northern leaf blight (NLB), a common foliar disease of maize, were annotated in each image by one of two human experts. The three data sets together contain 18,222 images annotated with 105,705 NLB lesions, making this the largest publicly available image set annotated for a single plant disease.

From visual estimates to fully automated sensor-based measurements of plant disease severity: status and challenges for improving accuracy

“…4). Many images are needed for deep learning systems (Wiesner-Hanks et al 2018;Ramcharan et al 2019). The largest database for a disease was reported in 2018 (Wiesner-Hanks et al 2018); there were 8222 images of corn leaves annotated with 105,705 lesions of northern leaf blight, although all were from a single field in New York.…”

Millimeter-Level Plant Disease Detection From Aerial Photographs via Deep Learning and Crowdsourced Data

“…A chief limitation of this method is the difficulty of acquiring field images at high enough resolution and clarity such that individual lesions can be discerned. Capturing images in which each pixel represented a millimeter or less at canopy level required slow flights at low altitude with a high-zoom lens (Wiesner-Hanks et al, 2018), not ideal for comprehensively imaging a large area. This challenge would be even greater when working with a disease with small or inconspicuous symptoms-chlorosis, leaf curling, lesions only a few millimeters in diameter-as opposed to the large, obvious lesions of NLB.…”

“…All Mechanical Turk human intelligence tasks (HITs) consisted of one or more prompts to draw a single bounding polygon delineating the boundaries of a single lesion (Figure 2, top right), previously annotated with a line down the major axis by one of two human experts (Wiesner-Hanks et al, 2018). All images and annotations used, generated, or described herein are available in an Open Science Framework repository (https://osf.io/p67rz).…”

“…The IoU was also used to filter out low-quality lesions. Entire UAV images were filtered automatically prior to annotation and manually during annotation, as described previously (Wiesner-Hanks et al, 2018), but individual lesions in an image could still vary in how clear and defined the visible symptoms were. Preliminary manual inspection of MTurk annotations revealed that lesions on which otherwise well-performing workers drew lesions with low overlap with one another were often blurry, ambiguous, or otherwise unacceptable.…”

“…In the course of our previous work, we labeled over 100,000 examples of northern leaf blight (NLB), a fungal foliar disease of maize that causes gray-brown necrotic lesions (Wiesner-Hanks et al, 2018). Each of these annotations consisted of a line drawn down the principal axis of a lesion.…”