Brown spot, bacterial leaf blight and rice blast are diseases of rice that reduce productivity. Even if only a few rice plants in a field are infected, these diseases can spread fast and destroy the whole crop.
Brown spot, bacterial leaf blight and rice blast. Images by Donald Groth and William M Jr via Wikimedia Commons
These diseases can be distinguished by patterns they form on the leaf. Can we create a system to automatically detect these patterns and, therefore, the diseases?
Recently, researchers from Tamil Nadu and Karnataka came up with a method to do this. They went around Neduvasal in Pudukkottai district, Tamil Nadu, and, using a Canon D3000 digital camera, they collected around 125 images of rice plants with and without these diseases.
Digital images often contain artefacts. The researchers removed speckles and blurs from the images using a Wiener filter.
Now, the problem was to remove the background and retain only the leaves. The researchers used the K-means clustering algorithm, which partitions data into clusters based on centroids. The centroids in the background create clusters that are distinct from those in the leaves – a simple unsupervised method that separates the leaves from the background.
Leaves with brown spot, bacterial leaf blight and rice blast have distinct shapes and colours. So the researchers extracted colour features from the images of the leaves using a colour concurrence matrix, a pixel colour probability for colour features. Using the area and diameter of the infected region of the leaves, they extracted shapes, and texture features using grey levels and leaf edges.
The researchers used the colour, shape and texture extracted to train a combination of support vector machine and a neural network algorithm.
‘A combination model is faster to train and more accurate,’ explains K S Archana, Vels Institute of Science, Technology and Advanced Studies, Chennai.
The researchers tested their model and compared the results with results from existing algorithms. Their model had better accuracy.
The accuracy ranged from 97% to 99% for distinguishing rice blast and brown spot from healthy leaves. For bacterial leaf blight, it was around 95%.
Imagine. Soon, farmers may be able to use a drone to click pictures of rice plants in their fields to identify unhealthy plants and use timely targeted treatments.
What is more interesting is that none of the researchers in the team had an agriculture background. The team consisted mostly of engineers, some technology developers and one biomedical researcher. But now they require agriculture scientists to test their system extensively. Any takers?
The Journal of Supercomputing, 22 January 2022;
National Remote Sensing Centre
STEAMindiaReports: science news free to use for Indian media