Deep learning is used for coffee disease classification

The largest coffee producer and exporter in the world is Brazil. This study by João Vitor Yukio Bordin Yamashita, and João Paulo R.R. Leite identifies coffee diseases on lea es on low cost coffee farms in loco without internet connection. In order to achieve production quality and productivity then the coffee diseases on the leaves needs to be detected early. In order to detect the diseases two data were used and were in the form of images with over 600 images used. The results showed that the diseases were mostly from coffee farms that are not exposed to a lot of resources.

About coffee leaf diseases

A task that is known to be difficult is detecting diseases in coffee plants as a variety of pests can affect them. Further, coffee leaves are analysed by human experts, however this process is very expensive and time consuming. The important aspects is treating the coffee disease at a very early stage which is efficient to get the right quality of beans. This process is a disadvantage to small coffee producers as they can not afford to hire a coffee production expert. The international coffee organization has mentioned that Brazil is the world’s largest coffee producer. Further, statistics show that “information collected by the CNA (Brazilian Agriculture and Livestock Confederation) confirms the contribution of small properties to coffee production: of all coffee farms in the country, 88% have an area of less than 50 hectares. The majority (69%) have an area between 2 and 20 hectare.”

The detection of coffee leaf disease

Machine learning has been a solution in detecting the coffee leaf diseases issue. Further to have better results, onvolutional neural networks (CNN) and deep learning models have been used. The aim and the objective of the study is as follows: “the aim of this study is to embed convolutional networks in a low-cost microcontrolled board to automatically classify diseases on coffee leaves without the need for an internet connection and test it in a real-world scenario. This solution would especially benefit small farmers, who cannot afford more sophisticated resources and human

Expertise.”

The procedures in the study

The study was analysed in three procedures: 

1. “Preprocessing and using techniques such as data augmentation”

2. “Training the CNN deep learning model, using transfer learning and powerful GPUs”

3. “Deployment, using optimisation and quantisation techniques to reduce the memory and processing power needed to run the model. All of them have been developed and improved by the scientific community in recent years.”

Methods and techniques used to detect the coffee leaf disease

With CNN the biggest problem in analysing the diseases is in having insufficient data both in diversity and quantity. When there is a challenge in quality then data augmentation can be put in place. Further, techniques of machine learning have been used to address the coffee leaf disease issue. There were other techniques used to detect the disease such as TensorFlow Lite, MobileNet architecture, development board, datasets and proposed architects. In addition, data augmentation, train validation, test sets and a training framework. Others include: model training, evaluation metrics, single-stage model, cascade model, onboard performance evaluation, usability and affordability. 

Conclusion of the study

The study mentioned that convolutional neural networks to detect the diseases on coffee leaves was verified and does not require an internet connection. This method can benefit low cost coffee producers that have less resources. The models showed the following results, “almost all performance indices were increased in the cascade model in relation to those from the single-stage model, except for those related to Rust class, which had a small decrease. The single- stage model showed difficulties in diseases with more subtle or similar characteristics, such as Cercospora, Phoma and Leaf Miner.”  The results show that the datasets need to be improved as well as procedure of the onboard system. It has been mentioned that some coffee plants might be affected by more than one disease. More research needs to be out taken in this sector.