Advances in agri-tech allow coffee farmers to instantly identify pests and diseases for quick treatment and prevention in the field. Photo credit: Bhavi Patel
Coffee is grown in more than 70 countries, providing livelihoods to millions of people. However, every coffee producer faces countless daily challenges–the effects of climate change, labor shortage, market uncertainties, pests and diseases, and many more. Leaf diseases like coffee leaf rust or Roya, which are caused by the fungal pathogen Hemileia vastatrix are some of the most devastating and difficult to overcome. As history has shown, if not detected early and effectively managed, coffee leaf rust can cause widespread crop loss, significantly affecting the coffee producer and even the global coffee supply chain.
In light of the deadly outcomes of a Roya infection on a coffee farm, any means to enable farmers to detect an outbreak as early as possible would be hugely helpful and advances in agri-tech could make this possible.
Two scientists, Opeyemi Adelaja and Bernardi Pranggono at the School of Computing and Information Science, Anglia Ruskin University, Cambridge, UK have conducted a study that aims to address the limitations of the existing methods and provide a more efficient, reliable, and cost-effective solution for coffee leaf disease identification. The study presents a novel approach to the real-time identification of coffee leaf diseases using deep learning. As part of their research, they implemented several transfer learning models, including ResNet101, Xception, CoffNet, and VGG16 to evaluate the feasibility and reliability of their solution. The experiment results show that the proposed models achieved high accuracy rates of 97.30%, 97.60%, 97.88%, and 99.89%, respectively. CoffNet, the preferred processing model, showed a notable processing speed of 125.93 frames per second (fps), making it suitable for real-time applications.
“Traditional methods, such as manual inspection by skilled workers or lab-based techniques like PCR, are labor-intensive, time-consuming, and often impractical for large-scale operations,” shares Dr. Pranggono. “In contrast, our deep learning approach, exemplified by CoffNet’s processing speed of 125.93 frames per second, allows farmers to detect diseases instantly using devices like smartphones or drones. This speed and accessibility reduce delays in diagnosis—potentially by days or weeks compared to traditional methods—minimizing disease spread and yield losses.”
“This shift to automated, real-time systems could significantly boost efficiency and crop health management,” he adds.
The study made use of a freely available BRACOL dataset. The leaves for the images used in the study were collected from Santa Maria of Marechal Floreano in Brazil’s hilly state of Espirit Santo and the images of the abaxial side (underside) of the leaves were captured using various widely available smartphones, reducing the need for specialized equipment. A comprehensive collection of 1747 images depicting Arabica coffee leaves was compiled, featuring both healthy specimens and those affected by various types of biotic stress. These biotic stresses were identified by an expert who analyzed the images and ensured that the dataset was accurately labeled based on the main biotic stress affecting each leaf and its severity level.
“By simplifying disease detection and reducing workload, this technology empowers these farmers to compete with larger operations, promoting social equity and sustainable practices,” explains Dr. Pranggono.
The study demonstrates a promising proof of concept with CoffNet’s high accuracy (97.88%) and real-time capabilities. The technology has the potential to revolutionize coffee leaf disease identification and management, but there is still a long way to go before large-scale adoption takes place.
“Real-time identification and early detection, if it can be implemented on a large scale, will help farmers in carrying out preventive spray before the spread of the pathogen,” shares Rohan Kuriyan, coffee producer at Balanoor Plantations in India. “This will definitely help in salvaging the crop, which due to challenges like climate change, is dropping at alarming levels.”
Integrating the deep learning model with IoT devices like sensors and drones could enable automated, field-wide monitoring. Developing intuitive mobile apps or interfaces tested on real coffee farms by coffee growers would make it accessible to non-technical users. The dataset also needs to be expanded to include diverse real-world conditions, enhancing the robustness of the model. The tech would also need the support of global agricultural organizations, national coffee governing bodies, and tech companies to take it mainstream.
With further research and testing, the technology also holds the potential to evolve from coffee to other crops, amplifying its impact on global agriculture. Dr. Pranggono emphasizes that collaboration with the farmers is key. “We are not just building tools but aiming to solve real-world problems they face daily,” he says.