Texture Classification of Paddy Soil using Convolutional Neural Network

Abstract

This research aims to classify the types of paddy soil based on its image, which consists of sand, clay and clay sand. The data is divided into 3 parts, training data of 1277 images and data testing and validation of 160 images with 224x224 pixel. Some of the transfer learning models used in this trial are Resnet152, VGG16, Xception, MobilenetV2 and DenseNet201, with 10 epochs and a batch size of 32. The results show that the accuracy of the Resnet152 model has 63.75% at best epoch 5, the VGG16 model 90.62% at epoch 9, the Xception model 91.25% at epoch 10, the MobilenetV2 model 86.87% at epoch 7 and Densenet201 model 93.12% at epoch 7. The model with the highest accuracy, Densenet201, was then tested with 100 epochs resulting in 95% accuracy. This research shows that the Densenet201 model is the best model for classifying paddy soil types.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint