ERFORMANCE ANALYSIS OF ALEXNET CONVOLUTIONAL NEURAL NETWORK (CNN) ARCHITECTURE WITH IMAGE OBJECTS OF RICE PLANT LEAVES

Abstract

Rice is a staple food consumed by Indonesian people, even 75% of the world's population consumes rice and it is mostly found in Asia. Rice derived from pounded rice is a staple food so it can be consumed. In the process of planting rice, pests and diseases are not spared so that it can affect crop yields. Pest and disease attacks need fast, accurate and precise handling so that crop failures. In this paper, we will discuss the classification of leaf diseases of rice plants using the Convolutional Neural Network (CNNN) algorithm, especially the Alexnet architecture. There are 4 types of disease, namely Brown spot, Leafblast, Hispa and Healthy. Models built based on the Alexnet architecture may have differences in the level of accuracy and loss compared to other architectures due to the different stages in the sequential model formation. The dataset used is public data from Kaggle consisting of 4 classes with a total of 1,600 images. In each class the dataset is divided for training, testing and validation datasets with a ratio of 70:20:10. As for tools in the process of training datasets using Google Colab from Google. After going through the stages of the research, the research results obtained are accuracy worth 99,22%, mean average precision worth 0,24 and loss worth 0,05.