Development of convolution neural network model with resnet-50 for chili plant disease detection in android application

Authors

  • Hendra Bayu Suseno Universitas Islam Negeri Syarif Hidayatullah jakarta
  • Ilham fiqri Saputra Universitas Islam Negeri Syarif Hidayatullah Jakarta
  • Victor Amrizal Universitas Islam Negeri Syarif Hidayatullah Jakarta
  • Basiroh Universitas Islam Batik Surakarta
  • Feri Fahrianto Universitas Islam Negeri Syarif Hidayatullah Jakarta

Keywords:

Chili, Disease detection, Image Classification, ResNet50, TensorFlow Lite

Abstract

Chili is a strategic horticultural commodity in Indonesia with demand increasing every year. However, chili productivity still faces serious obstacles due to the attack of various leaf diseases such as curling, spotting, yellowing and aphids, which are the often not detected early because they still rely on manual visual identification methods. To answer these challenges, this research aims to design and develop system for leaf image-based detection of chili plant disease using the Convolutional Neural Network (CNN) algorithm with ResNet 50 architecture, which is integrated in an Android application. The methodology used in system development refers to the CRISP-DM approach, starting from problem understanding, image processing, model training, to implementation and evaluation. The dataset consists of five classes: healthy, leaf curl, leaf spot, yellowish, alphid infestation. The model was converted to TensorFlow Lite format and integrated into an Android application that was tested using black-box testing and user satisfaction evaluation through SUS questionnaire. The test results show that the system runs as expected, both from the technical and usability aspects. This system is expected to help farmers in conducting disease detection quickly, accurately, and efficiently, and become the first step towards the application of digital agriculture based on artificial intelligence. With 25 epochs of training, the results show a model accuracy of 93.20% and validation accuracy of 91.80%, with an average confidence level above 90% on test images.

Author Biographies

Ilham fiqri Saputra, Universitas Islam Negeri Syarif Hidayatullah Jakarta

Departement of Informatics

Victor Amrizal , Universitas Islam Negeri Syarif Hidayatullah Jakarta

Departement InformaticsĀ 

Feri Fahrianto, Universitas Islam Negeri Syarif Hidayatullah Jakarta

Departement Informatics

Downloads

Published

2026-10-31

Issue

Section

Articles