Mobile app uses artificial intelligence to diagnose crop diseases

Mobile app uses artificial intelligence to diagnose crop diseases

Researchers have developed a mobile app that diagnoses cassava diseases in the field using images captured on a smartphone, helping farmers to identify and manage diseases and reduce losses.

Pests and diseases reduce cassava yields in Africa by more than half. Plant health problems include cassava brown streak disease (CBSD), cassava mosaic disease (CMD), brown leaf spot (BLS), cassava green mites (Mononychellus tanajoa) and cassava red spider mites (Oligonychus biharensis). In fact, mites are so small that they are readily confused with plant diseases. Early identification of these plant health problems would help farmers to manage them and reduce significant losses.

Babuali Ahmed, a researcher at the International Institute of Tropical Agriculture (IITA) in Tanzania, explains that to diagnose these pests and diseases, “we can use our eyes, but that means we have to be well trained and experienced. We can go to the lab, but this is expensive and there are not many labs out there. Of course, we can take pictures with our smart phones and send them to experts, or even better, what if we could train our smart phone to diagnose the diseases for us, from the images they capture?”

Nuru’s database contains more than 15,000 images of cassava leaflets. Credit: E.Massam/IITA

That is what researchers at IITA and Penn State University have done; they have created an app that identifies cassava diseases and pests. The app is called ‘Nuru’ which means ‘light’ in Swahili. It was designed in collaboration with Google, using TensorFlow, an open source software for object recognition. Nuru was crafted by taking 11,670 photos, to create 2,756 images of leaves which were cropped to form 15,000 images of leaflets. After experts diagnosed the diseases, the photos were organized into a database which was used to train the software using machine learning to recognize the symptoms of cassava pests and diseases.

The app is user-friendly, and farmers or extension agents simply point their smartphone camera at several cassava leaves and Nuru responds with a diagnosis. The app was purposefully designed to be cautious, so if it doesn’t recognize the disease, it gives no diagnosis. It can also be used offline, which is important in remote farm communities.

The IITA-Tanzania team working to validate the app in real farming conditions in sub-Saharan Africa has tested its ability to correctly diagnose disease symptoms. It is correct roughly 70-80% of the time, which while not quite as good as experienced cassava researchers, is already better than trained extension workers or farmers. Furthermore, as its development continues, it will keep getting better.

“Smartphones are becoming more and more common in rural Africa. Smallholders or extension officers with a basic smartphone with a camera can already download the app for free via Google’s Play Store, fire it up, point it at a leaf with disease symptoms and get an instant diagnosis. That is truly revolutionary!” said James Legg, a plant virologist at IITA and RTB’s Flagship Project 3 leader. Furthermore, reports across the globe are picked up in real-time by the team of cassava disease experts involved in Nuru’s development, which allows direct interaction between users and research specialists. Although Nuru provides advice on how to manage each of the major cassava diseases identified, this direct link with researchers means that users can also get personalized advice appropriate to their situation. Data generated through the app can also be used to identify new areas of disease occurrence and to generate maps to plot disease spread, which helps to target control interventions.

The team that designed the app was awarded a USD100,000 prize for their creation by the CGIAR Platform for Big Data in Agriculture during their annual conference in in 2017. This is helping to fund the development and validation of the app. Efforts in 2018 will focus on extending the AI technology used to the other RTB crops (including banana, potato and sweet potato), and on providing training on app usage to farming communities in Kenya and Tanzania, including 28,000 farmers working with Self-Help Africa. Data generated by the app will also be used to plot disease incidence and spread.

Photo: The user-friendly app allows farmers and extension agents to diagnose diseases in the field. Credit: IITA