A team of scientists working to reduce the impact of a devastating banana virus in Burundi has developed an Artificial Intelligence (AI) tool to control the disease.
Research by the Food and Agriculture Organization of the United Nations (FAO) shows that Banana Bunchy Top Disease (BBTD), caused by the Banana Bunchy Top Virus (BBTV), is endemic in many banana-producing countries in sub-Saharan Africa.
The virus was first reported in the Democratic Republic of Congo (DRC) in the 1950s and has become invasive and spread to 15 countries in sub-Saharan Africa. The disease has been reported in Angola, Benin, Burundi, Cameroon, Central African Republic, Republic of Congo, DRC, Equatorial Guinea, Gabon, Malawi, Mozambique, Nigeria, Rwanda, South Africa and Zambia. Recent findings, however, show that BBTD is now a major threat to banana cultivation and a threat to the more than 100 million people who rely on bananas as their staple food.
The AI development team, co-led by Dr. Guy Blomme and Dr. Michael Gomez Selvaraj of the Alliance of Bioversity and CIAT (ABC), tested the detection of banana plants and their main diseases using aerial imagery and machine learning methods.
This project aimed to develop an AI-based banana disease and pest detection system using Deep Convolutional Neural Network (DCNN) to support banana farmers. As farmers struggle to defend their crops from pests, ABC scientists have created an easy-to-use tool to detect banana pests and diseases.
The tool, which has proven to be 90% successful in detection in some countries such as the DRC and Uganda, is an important step towards creating a satellite-powered network to monitor disease and pest outbreaks, the researchers say.
In the test phase, in collaboration with a team from the national agricultural research organization of Burundi – ISABU, two areas in Cibitoke province where banana bunchy top disease is endemic were compared with a disease-free area in Gitega province (Central). Cibitoke province is endemic for BBTD and is located in a border area bordering Rwanda and the Democratic Republic of the Congo (DRC).
Performance and validation measures have also been calculated to measure the accuracy of different models in automated disease detection methods by applying state-of-the-art deep learning techniques to detect disease and pest symptoms individually in different parts of the plant.
The researchers explained the reasons why disease detection in bananas is essential.
“In East and Central Africa, it is an important dietary component, accounting for more than 50% of total daily food intake in parts of Uganda and Rwanda.”
Banana is also dominant in Burundi. The cultivated area is estimated to be between 200,000 and 300,000 ha, between 20 and 30% of agricultural land. Data from Burundi’s Ministry of Agriculture and Livestock indicate that food security and nutrition continue to deteriorate, with 21% of the population food insecure.
They say this can be exacerbated by various plant diseases such as BBTD. While bananas are critical to people’s food security and livelihoods, experts also say BBTD could have a devastating economic and social impact on the continent.
“When BBTD enters, it is initially a very cryptic disease and shows no visible symptoms,” said Bonaventure Omondi, a CGIAR researcher who collaborated on this project and works on banana diseases and seed systems projects. In an IPS interview. While stopping the disease early was crucial, it was also a challenge, which is why an AI solution was essential.
According to agricultural experts, the East African Highlands is the site of secondary diversity of a type of banana called the AAA-EA type. These bananas are genetically close to dessert bananas, but have been selected for use as brewing, cooking and dessert bananas.
Banana cultivation in Burundi falls into three different categories. Bananas that are juiced and fermented for beer/wine account for about 77% of national production by volume. 14% of bananas are grown for cooking, and finally about five percent are dessert bananas, which are ripe and eaten directly. With recent advances in machine learning, researchers were convinced that new disease diagnosis based on automated image recognition was technically feasible.
“Minimizing the effects of disease threats and maintaining a mixed matrix of banana and non-banana fat is a key step in managing a wide range of diseases and pests,” Omondi said.
As an example of how this emerging technology works, researchers rely on datasets imaged from banana crops showing disease symptoms and established algorithms to help identify diseased plantations.
Prosper Ntirampeba, a banana grower in Cibitoke province in northwestern Burundi, told IPS that he harvested fewer bunches of bananas last season because of the spread of BBTD across his farmland.
“Since this disease arrived in our main production area we have been forced to uproot the infected plants. This led to a heavy burden of additional costs,” he said.
In another case, after BBTD was detected, agricultural officials under the command of the researchers advised farmers to remove all infected “mats”, where several hectares of diseased plants were destroyed. That is the key to eliminating the disease in Busoni, a remote rural village in northern Burundi. Although some farmers often resist removing banana plants, Ntirampeba said it is essential to get rid of the disease.
“The disease is likely to threaten the livelihood of most farmers who depend on the crop,” he told IPS.
Other new disease surveillance methods are also currently being developed in Burundi by ABC researchers, including drone surveillance to determine local disease risk and delimit recovery areas.