Classification of Soil Fertility Level Based on Texture with Convolutional Neural Network (CNN) Algorithm

Abstract

This study aims to classify the level of agricultural soil fertility based on texture. This study proposes a new approach to categorizing soil fertility levels based on soil texture using the Convolutional Neural Network (CNN) algorithm. The data is divided into three parts, with a percentage of 80% training data for 1120 images and 10% for each test and validation data for 140 soil images with a data size of 224x224 pixels. Several trials have been conducted by tuning the learning rate, optimizer, batch size, and augmentation. The results show that using the Adam optimizer, learning rate 0.001, batch size 8, and augmenting the dataset produces the best accuracy of 94.24% at best epoch 73. This research shows that by tuning the CNN parameter, it can classify the level of agricultural soil fertility based on its texture.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint