2017 WinnerPest and disease monitoring by using artificial intelligence
The Inspire Challenge is an initiative to challenge partners, universities, and others to use CGIAR data to create innovative pilot projects that will scale. We look for novel approaches that democratize data-driven insights to inform local, national, regional, and global policies and applications in agriculture and food security in real time; helping people–especially smallholder farmers and producers–to lead happier and healthier lives.
This proposal was selected as a 2017 winner, with the team receiving 100,000 USD to put their ideas into practice.
Pest and disease monitoring by using artificial intelligence
The project expects to radically transform pest and disease monitoring by using artificial intelligence (AI), advanced sensor technology and crowdsourcing capable of connecting the global agricultural community to help smallholder farmers. It aims to increase the effectiveness of farm-level advice by leveraging three critical advances:
- The democratization of AI thanks to open access platforms such as Google’s TensorFlow.
- The miniaturization of technology allowing affordable deployment.
- The development of massive communication and money exchange platforms such as M-Pesa that allow rural extension to scale as a viable economic model enabling last mile delivery in local languages.
More than 200,000 images of diseased crops have already been collected on farms at RTB cassava field sites in coastal Tanzania and western Kenya using cameras, spectrophotometers, and drones. These images will be used to train AI algorithms.
The team recently developed an AI algorithm with TensorFlow that can automatically classify five cassava diseases building on published work covering 26 diseases in 12 crops. Through a collaboration with Google, a TensorFlow smartphone app has been developed and field-tested in Tanzania by IITA.
Step by step
The project was one of five winners of the Inspire Challenge 2017 and was awarded US$100K at the inaugural annual convention of the CGIAR Platform Big Data in Agriculture, during 19-22 of September.
Developing a mobile AI assistant that diagnoses cassava diseases
The team successfully developed a mobile AI assistant that works inside a standard smartphone and is capable of accurately diagnosing cassava diseases offline, without an internet connection. The assistant is called Nuru, which means ‘light’ in Swahili.
Under field conditions, Nuru was, on average, twice as accurate as the extension workers she was tested against. Nuru is linked to PlantVillage and allows advice from experts (at CGIAR/FAO/governments) to be sent offline and in local languages (currently in Swahili, French, Twi, Hindi, and English).
Training the convolutional neural network
The AI relies upon TensorFlow, a machine learning environment where a convolutional neural network (CNN) is trained to recognize crop diseases based on images collected at IITA research plots. The project overcame a number of significant challenges in getting an accurate CNN to work offline inside an Android app; in-phone CNN deployment is not yet a standard approach and most systems rely upon the cloud, which would not be functional in a smallholder farmer setting where connectivity is poor. The CNN is a work in progress and will improve over time with more training.
Building an expert portal on the PlantVillage platform
The project also built an expert portal on the PlantVillage platform for IITA scientists to examine records of diseases diagnosed by Nuru. Experts are automatically notified when a user receives a diagnosis, and they are shown the images and data (location, AI accuracy, time) associated with the diagnosis. This allows IITA to both check the accuracy of the AI assistant and note where and when diseases are being recorded.
An example of the portal in action was the identification of Cassava Mosaic Disease (CMD) in Oddar Meanchey Province, Cambodia where there had previously been only unconfirmed reports. The expert portal enabled IITA’s James Legg to interact directly with the Nuru user in Cambodia and discuss monitoring and control of CMD. This was a clear demonstration of Nuru’s potential to provide early warning support for the identification and management of new crop disease outbreaks.
AI takes root
The team annotated thousands of cassava plant images, identifying and classifying diseases to train a machine learning model using TensorFlow. Once the model was trained to identify diseases, it was deployed in the app. Farmers can wave their phone in front of a cassava leaf and if a plant had a disease, the app could identify it and give options on the best ways to manage it.
For more information on how Nuru is helping farmers better identify and manage diseases quickly, watch Google’s video spotlight on the project below and read more on Google’s blog, The Keyword.
Expansion to other crop diseases
While the project focused on cassava disease detection, it also trained an accurate model for potato diseases which has been field tested and will be deployed in India in October 2018 with 1,000 farmers as part of the Indian Government’s FarmerZone platform. The application is available in Hindi and Punjabi and will scale up its reach to 200,000 farmers within 24 months.
This benefited from CGIAR experts in CIP. The team are also collecting images of banana and sweet potato diseases. In collaboration with the United Nations Food and Agricultural Organization (FAO), the team also developed a tool to identify fall armyworm damage.
US$250K scale-up grant
The project was awarded a 2019 Inspire Challenge Scale-up grant of US$250K at the third annual convention of the CGIAR Platform for Big Data in Agriculture, 16-18 October 2019.
Nuru AI was made freely-available for download onto Android devices through Google Play at the end of June 2018. It has been downloaded by users on all continents and used extensively in Africa and South-East Asia. This is the first stage of the application roll-out; additional funding would allow roll-out at a broader scale.