Classification method of applying types of rice fertilizers using Resnet50 architecture

Abstract

The Indonesian government has implemented various strategies to increase rice production and productivity. However, until now, the results have not met expectations, and the sustainability of rice farming practices in Indonesia is still poor. One of the important problems that needs to be addressed is the imbalance of fertilizer use, as it can cause various problems in rice cultivation that lead to non-optimal productivity of rice plants, such as reduced yields and decreased quality of rice grains. Various techniques have been developed to determine the appropriate fertilizer for rice plants based on leaf color of their leaves. However, using specific algorithms to solve the illumination problem increases the computational process and still leaves the possibility of inaccurate image representation. In addition, the use of UAVs is very expensive, making their implementation difficult for farmers. Beside that, previous studies generally only classify Nitrogen status into low or high, fertile or infertile classes, whereas each fertilizer has different characteristics. Therefore, this study proposes a classification method of applying types of rice fertilizers based on vegetative microscopic images of rice leaves using the ResNet50 architecture. The proposed method uses Resnet50 architecture of Convolutional Neural Network to analyze microscopic rice leaf images and classify three types of rice fertilizers accurately, quickly and non-destructively.