← All solutions

Agriculture

Banana disease detection tool

The AI solution aim to solve the early detection of Black Sigatoka and Fusarium Wilt in Banana

Agriculture Sector Unit · Coordinator: Dr. Devotha Nyambo

The problem

Problem

The AI solution aim to solve the early detection of Black Sigatoka and Fusarium Wilt in Banana

Background

Diseases like Fusarium Wilt and Black Sigatoka affect the banana plant and lead to a decrease in plant yields. Most local banana farmers use simple naked-eye observation done by the farmers themselves and the agricultural experts to detect the presence of diseases on the plants, which is cumbersome, costly, prone to errors, and time-consuming. The current state of plant disease diagnosis is transitioning from disease identification using visible symptoms by the eyes to the use of automatic, data-driven solutions applying deep learning and computer vision techniques, which are easier and cheaper. 

Consequences

If the problem is not addressed crucial management steps will not be taken to mitigate or avoid the effects of Black Sigatoka and Fusarium Wilt banana diseases early and this will result in great yield losses between 30% to 100% of banana fruits.

Specific issue addressed

The AI solution addresses the early detection of Black Sigatoka and Fusarium Wilt banana diseases from their symptoms seen on the images of banana leaves and stalks.

Signs and indicators

The symptoms of Black Sigatoka banana disease are small, pale-yellow spots on young banana leaves, which enlarge, turn brown, and have light gray centers within days with yellowed surrounding tissue. 

The symptoms of Fusarium Wilt include wilting and yellowing older leaves, followed by new ones. As the disease progresses, the yellowed leaves collapse, forming a dead leaf covering around the pseudostem. This continues until all leaves fall and dry up, leading to a dead plant. Splitting of the pseudostem’s base is another common sign. 

How it is handled today

Most local banana farmers use simple naked-eye observation done by the farmers themselves and the agricultural experts to detect the presence of diseases on the plants, which is cumbersome, costly, prone to errors, and time-consuming. 

What the tool improves

The AI tool aim to improve outcomes by automating the detection of Black Sigatoka and Fusarium Wilt banana diseases.

Scale and context

Global picture

Based on the report by the Food and Agricultural Organization Statistics, in 2021 the annual global production of bananas was 124 million tons, and 54.4% of these tones were produced in Asia, with India leading by producing 33 million tons and Africa producing 22 million tons which is equivalent to 22.7% of the global production.

In Tanzania and the region

In Tanzania, a total of 20 735 banana-growing smallholder farmers were reported to have banana disease occurrences in 2011.

Projections

In Tanzania, a total of 20 735 banana-growing smallholder farmers were reported to have banana disease occurrences in 2011.

How it works

Purpose

The purpose of the AI tool is to automate the early detection of Black Sigatoka and Fusarium Wilt banana diseases. 

Origin

The AI tool was developed as part of a university research.

Technology

Deep learning technology is used in the development of the tool.

How it works, step by step

The AI tool is a mobile application that can capture an image using the phone camera or upload images from the gallery. After the image is captured or uploaded it is sent to the server to be detected by the model if it is a healthy banana leaf image, an image of a banana leaf affected by Black Sigatoka disease, an image of a banana leaf or stalk affected by Fusarium Wilt disease, or if the image is not of the banana plant. After a few seconds, the result is sent from the server to the mobile application with the class of the image and the confidence score of the model’s prediction. From these results the farmer will be informed if his banana plant is healthy or if it has one of the two diseases. Then if a disease is detected recommendations for mitigating the effects of the diseases on the farms are provided. 

From the application the farmer can read information about banana, about Black Sigatoka and Fusarium Wilt banana diseases and the farmer can also change the language between English and Swahili which are supported by the application. 

Data

Data sources

Images of banana leaves and stalks collected from the farms in Arusha, Mbeya, Kagera, Kilimanjaro, and Dar es Salaam regions of Tanzania were used to train the AI tool.

Data quality and validation

The data was annotated using LabelMe software. 

Inputs

The input data required are the images of banana leaves or stalks.

Outputs

The output provided by the tool is the predicted class and the model’s prediction confidence score.

Who it helps

Primary users

Farmers are the primary users of the tool.

Recommendations it gives

Yes. The tool provides recommendations for mitigating the effects of Black Sigatoka and Fusarium Wilt banana diseases. For example fungicides and ways to avoid the spreading to the diseases.

Efficiency gains

The AI tool improve efficiency compared to traditional methods by making prediction bases on experiences or examples of tens of thousands of labelled images. The AI tool is also cheaper compared to the costs of hiring agricultural experts. 

Time saved

The AI tool can detect the presence of absence of diseases in less than 5 seconds.

Improved access to services

The model was deployed in a server and can allow concurrent users from the mobile applications.

Evaluation and limitations

Accuracy

The AI tool is not always accurate

Validation feedback

Not available

Testing

No recommendations

Current limitations

The AI tool is not always accurate

Technical challenges

Images of plants with similar symptoms from different causes may be predicted to have one of the two detected diseases.

Handling discrepancies

No discrepancies exist

Performance monitoring

By the number of correct predictions it provides compared to the decisions of agricultural experts

Evaluation questions

How can the AI tool’s accuracy be improved?

Impact

Practical applications

Practical applications of the AI tool are

Disease detection and mitigation

Support for decision-making

By detecting Fusarium Wilt banana disease, the tool can support decisions to start plantation in a new field because Fusarium Wilt is caused by a soil-borne fungus

By diagnosing many banana plants in an area using the tool, decisions can be made about the extent of the disease in an area and what collective measures can be taken to mitigate the effects of those diseases

Community benefits

Benefits offered by the AI solution to target communities are:

Cost reduction

Ease of use

Scaling opportunities

Yes, there are scalability opportunities for the tool across other regions and other plant diseases

Next steps

Long-term goals

There are no long-term goals for the AI tool.

Planned features

There are no additional features or upgrades planned.

Partnerships

Partnerships or collaborations can help to improve the tool and increase its accuracy hence enhancing its impact.

Access

Demonstration

Yes

Demo access

The application is not in play store yet. 

User guidelines

No there are no guidelines for documentation provided for using the tool.

Partners and funders

Implementation partners include TMDA, AGCOT and the EAC AI Alliance.