Use of deep learning for disease recognition in banana cultivation in the municipality of Paragominas/pa

Abstract

The state of Pará has stood out since the 1990s among the five largest national banana producers. In the northern region, the state of Pará stands out for being the one that produces the most bananas at a regional level. In 2020 Paragominas produced 416 tons of fruit, in an area designated for harvesting and a harvested area of 32 ha, with an average yield of 13,000 kg/ha. However, the occurrence of many diseases harmful to plantations ends up negatively interfering with production. The objective of this research is to develop a computational solution based on the machine learning technique, using digital image processing to automatically diagnose banana diseases. The images were captured on Paragominense soil in banana orchards infected by Black Sigatoka and Yellow Sigatoka. The technique known as data augmentation was applied to automatically generate more images of the classes. A Convolutional Neural Network was trained and subjected to test and validation data. The results showed that the Convolutional Neural Network was a robust and easily deployable strategy for detecting banana diseases. In the recent past, I hoped to bring a kind of revolution in agriculture and today these proofs are being confirmed with the numerous technologies available today. This significant high success rate makes the model a useful early disease detection tool in banana farming, and this research can be extended further to develop a fully automated mobile application to help banana producers locally, nationally and internationally.