Artificial intelligence is a rapidly developing technology taking the world by storm. Nowadays it’s hard to avoid news of some AI-related development making waves in the scientific community. Professors at universities everywhere are sounding the alarm over how students may be using large language models (LLM) like ChatGPT to sail through their courses without learning anything.
But such dark tales are balanced out in equal measure by hopeful signs. For instance, a report from China tells how researchers there are helping a paralyzed man regain some function in his lower limbs thanks to AI-powered implants in his brain. Here at Grow Further, we are closely following a project our donors are sponsoring that aims to unlock the power of machine learning, a subset of AI technology, to dramatically improve smallholder farmers’ yields.
In Arusha, Tanzania, computer science experts at the Nelson Mandela African Institution of Science and Technology (NM-AIST) and the Tanzania Agricultural Research Institute are developing a machine learning-driven smartphone application that’s smart enough to tell farmers when their crops are infected with diseases or threatened by pests. More often than not, farmers notice these threats too late. Entire fields are lost if interventions don’t come soon enough, imperiling food security. That’s why Dr. Neema Mduma and her co-researchers at NM-AIST are building a smartphone app that will alert farmers to these threats by analyzing photographs farmers take of their maize and common bean plants.
NM-AIST recently provided Grow Further donors with an update on their progress. We’re happy to report to our newsletter readers that they have made huge strides and are closing in on the prize. Once completed, the app will be made available to farmers and government ag extension workers throughout East Africa.
A treasure trove of data
Our partners in Tanzania have achieved two important milestones.
First, their efforts have already reached 1,500 farmers in the Arusha, Kilimanjaro, and Manyara regions. When our founder and CEO Peter Kelly visited NM-AIST on a due diligence mission, he noticed how many farmers’ maize crops were already badly impacted by pests and diseases. Peter was only able to inspect a handful of farms on that trip. That NM-AIST has since managed to gather data from more than 1,500 farms is a true achievement given how relatively young their research and development effort is.
Second, the data set they’ve collected from these 1,500+ farmers is truly massive. NM-AIST reports that they now have taken more than 500,000 images of crops in varying stages of health. The data sets include more than 254,000 separate images of maize plants, over 155,000 impressions of common bean plants, and nearly 9,000 images showing the progression and health status of other types of crops. They’ve also run quality checks on this data with the assistance of government agricultural extension officers and plant pathologists who can characterize various signs of pest infestations and disease infections.
The researchers said they’ve also employed computer algorithms to assist with their data quality control efforts. This is critical because the computer system they are developing must learn how to interpret photos showing signs of disease and pests so it can alert farmers to these problems quickly. The NM-AIST team members said their data quality control initiative partly entails cleaning up the data received, which means they must “identify and remove blurry, corrupted, or irrelevant images” as they said in their latest report.
A somewhat bumpy road
NM-AIST said its work connecting with over 1,500 farmers to generate more than 500,000 pieces of data on crop disease signs didn’t always go smoothly.
In some cases, the crops they inspected were too healthy, with the researchers noting that it proved “difficult to find diseased crops during the data collection in some areas.” The weather didn’t always cooperate, either. Ideally, conditions outdoors should be mild and sunny to allow for the best possible images of plants to be taken. Thus, they noted that some of their data collection efforts were hindered by heavy rainfall and excessively cloudy days.
Logistics is an ongoing challenge, too, as NM-AIST works to reach even more farms. Rural Tanzania is not entirely connected with the best road infrastructure. Sometimes, the greatest challenge is simply getting to the farmers where new potential data can be found.
Nevertheless, the team members have worked through these and other challenges. Their progress to date has been impressive.
They’ve completed site preparations and recruited enough farmers to see the project through. They’ve communicated openly and transparently with community members and stakeholders, fully disclosing what it is they are trying to do, how it will benefit Tanzania’s smallholder bean and maize farmers, and how the farmers and other stakeholders can help. Their data collection tool is working nicely, and they’ve trained ag extension officers and farmers on how to use it to help speed up data gathering—thus, the over 500,000 images collected in a relatively short period of time. And they say the initial data quality check is pretty much completed.
Onward and forward
Now, NM-AIST is busy cleaning up and properly labeling their data. Given the huge number of images collected, this will take them some time, but they will eventually end up with a complete set of annotated images that will form the foundation of their machine learning-based disease and pest detection application.
Next up, “preparation of document data sheets for image data set” and “uploading data to open access repositories,” two steps the researchers say they haven’t quite started yet but will soon. From the outset, NM-AIST pledged to make their data accessible to all through open-access repositories, so we’re happy to hear that they are upholding this promise, which will help to create more widespread social impact.
There is other detailed and sophisticated work left to be done, namely developing the machine learning model and deploying it to power the forthcoming smartphone application. Once it’s ready, they must train farmers on how to use the app and find a way to help spread this knowledge far and wide.
A final project evaluation is planned. We’ll be a part of that, and we are very much looking forward to congratulating NM-AIST on the successful launch of their innovative and potentially game-changing smartphone app. We can’t wait to see that made ready and available to smallholder maize and common bean farmers everywhere.
— Grow Further
Photo credit: Grow Further CEO Peter Kelly (left) inspecting a maize farm near NM-AIST’s headquarters in Arusha, Tanzania. True Vision Productions for Grow Further.