Deep convolutional neural networks (DCNNs) and transfer learning are being applied to just-in-time crop disease detection, through a smartphone tool that scans banana plants for signs of disease and pests. The tool is built into an app called Tumaini, which means “hope” in Swahili. Tumaini aims to link the farmer with extension workers who can help to stem an outbreak quickly. It can upload data to a global system for large-scale monitoring and control. The app’s goal is to facilitate a robust, easily deployable response to support banana farmers in need of crop disease control. Researchers from the International Center for Tropical Agriculture (CIAT), Biodiversity International, the Imayam Institute of Agriculture and Technology (IIAT), and Texas A&M University participated in the work. Rapid improvements in image-recognition technology made the Tumaini app possible. To build it, the researchers uploaded 20,000 images that depicted various visible banana disease and pest symptoms. They retrained three different convolutional neural network (CNN) architectures using a transfer learning approach. Six models were developed from 18 different classes using the images collected from different parts of the banana plant. Using artificial intelligence, scientists created an easy-to-use tool to detect banana diseases and pests. Courtesy of CIAT. In experiments, the team achieved an accuracy of about 90% in the detection rate. With a view toward running these detection capabilities on a mobile device, the researchers evaluated the performance of a single shot detector (MobileNetV1). They also computed performance and validation metrics to measure the accuracy of the different models in automated disease detection methods. Using a pretrained disease recognition model, the researchers were able to perform deep transfer learning to produce a network that could make accurate predictions. The Tumaini app scans photos of parts of the fruit, bunch, or plant to determine the nature of the disease or pest. It then provides the steps necessary to address the specific disease. It also records the data, including geographic location, and feeds it into a larger database. It can detect symptoms on any part of the crop and is trained to read low-quality images inclusive of background noise, such as other plants and leaves, to maximize accuracy. With an average 90% success rate in detecting a pest or a disease, the tool could help farmers avoid millions of dollars in losses. Courtesy of CIAT. “The overall high accuracy rates obtained while testing the beta version of the app show that Tumaini has what it takes to become a very useful early disease and pest detection tool,” said researcher Guy Blomme from Bioversity International. “It has great potential for eventual integration into a fully automated mobile app that integrates drone and satellite imagery to help millions of banana farmers in low-income countries have just-in-time access to information on crop diseases.” The banana mobile app is currently being tested by collaborative partners in Benin, the Democratic Republic of the Congo, Uganda, Colombia, China, and India. The researchers believe that the model system developed from this study could be transferable to other mandatory crops. A bunch of bananas grown by a smallholder farmer near Palmira in southwestern Colombia. Courtesy of CIAT/Neil Palmer. Future work will entail the development of a broad structure containing a trained model and an application for smartphone devices with features such as displaying recognized diseases in cassava, potato, and other crops. Additionally, future work will involve disseminating the use of the model by training it for banana disease recognition on wider applications, merging aerial images of banana-growing regions captured by drones and convolutional neural networks for instant segmentation of multiple diseases. By extending its research, the team hopes to make a positive impact on sustainable development and strengthen banana value chains, even creating a satellite-powered, globally connected network to control plant disease and pest outbreaks. “This is not just an app, but a tool that contributes to an early warning system that supports farmers directly, enabling better crop protection and development and decision-making to address food security,” said Michael Selvaraj, scientist at CIAT. The research was published in Plant Methods (https://doi.org/10.1186/s13007-019-0475-z). Tumaini is a mobile application that uses artificial intelligence to detect pests and diseases affecting bananas. The app works with just three steps: Take a picture of the plant, upload the image, and receive a specific diagnosis in a few seconds, along with control measures to aid in reducing yield losses. The app performs across various illuminations and resolutions to help smallholder farmers improve food production. For more information, visit phenomics.ciat.cgiar.org. Courtesy of CIAT.