According to a team of researchers from the Swiss Federal Institute of Technology (EPFL) in Lausanne, Switzerland, and Penn State, a computer network fed on a huge image dataset could indeed learn to identify specific diseases in plants with a high degree of accuracy, potentially opening the door for field-based crop-disease recognition using smartphones.
The researchers noted that producers in poor nations, including those in sub-Saharan Africa, who frequently lack the research infrastructure and agricultural extension programs to support smallholder farmers, may profit most from the technology. Plant diseases, which can lower yields or even eradicate a crop, are among the many variables that endanger global food security.
It has also been noted that plant diseases could have catastrophic effects on smallholder farmers, whose survival depends on producing healthy harvests. More than 80 percent of agricultural production in developing countries is produced by smallholder farmers, while up to 50 percent of hungry people reside in smallholder farming homes.
The first step to successful disease management is accurate disease identification. Disease diagnosis based on automatic image recognition, if technologically possible, might be made accessible on an unprecedented scale with the spread of smartphones and recent developments in machine learning and computer vision.
The researchers constructed a neural network, which is a sizable collection of computers with graphics processing units, to start testing this idea. They fed the network more than 53,000 photos of ailing and healthy plants to train it to identify patterns in the data using a deep-learning approach, a new branch of machine learning that employs algorithms to model high-level abstracts in data across several processing layers. The research draws on significant advancements made in computer vision over the previous few years, particularly in object detection.
Neural networks offer a mapping between an input, like an image of a sick plant, and an output, like a crop-disease pair. Deep neural networks have been used successfully across a wide range of industries. These networks are programmed by adjusting the network parameters so that throughout training, the mapping gets better.
Additionally, Facebook’s ability to recognize a user has been used by examining an uploaded photo. It is an illustration of how this new technology operates. These algorithms can categorize complex phenotypes, such as identifying a face, the author claimed. Researchers intend to utilize them to detect plant illnesses.
The photographs utilized in the study were a part of a collection of photographs that were accessible to the general public and were part of PlantVillage, a free online resource, and database. The data set showed 26 diseases as well as 14 crop species in both healthy and diseased states. Every image was given a class assignment, with each class denoting a crop-disease pair, and the researchers evaluated how well their model performed by classifying each image into the appropriate class. The classification work itself is fairly quick after the algorithms are constructed, and the final code is tiny enough to be readily installed on a smartphone. However, constructing the training and algorithms of the model demand a lot of computer power and time.
The technique has significant potential in a developed-world scenario in addition to helping producers in underdeveloped nations. This might be a resource for public institution land-grant extension staff as they support their grower clients and for the hordes of backyard gardeners who want to figure out what is causing their product to suffer. The strategy is designed to augment current disease identification methods, not replace them, the researchers noted in an online article published in Frontiers in Plant Science.
