Classification of Rice Leaf Diseases Using MobileNetV3-Large with a Grad-CAM-Based Explainable AI Approach on a Mobile Platform

Susanto, Adyatma Imam (2026) Classification of Rice Leaf Diseases Using MobileNetV3-Large with a Grad-CAM-Based Explainable AI Approach on a Mobile Platform. Undergraduate thesis, UPN Veteran Jawa Timur.

	Text (Cover) 22081010041.-cover.pdf Download (2MB)
	Text (Bab 1) 22081010041.-bab1.pdf Download (1MB)
	Text (Bab 2) 22081010041.-bab2.pdf Restricted to Repository staff only until 15 June 2098. Download (4MB)
	Text (Bab 3) 22081010041.-bab3.pdf Restricted to Repository staff only until 15 June 2029. Download (4MB)
	Text (Bab 4) 22081010041.-bab4.pdf Restricted to Repository staff only until 15 June 2028. Download (9MB)
	Text (Bab 5) 22081010041.-bab5.pdf Download (368kB)
	Text (Daftar Pustaka) 22081010041.-daftarpustaka.pdf Download (1MB)

Abstract

Rice productivity in Indonesia is frequently threatened by leaf diseases, which can cause yield losses. While Convolutional Neural Networks display high performance for disease classification, their "black box" nature often lacks transparency, making it hard for users to understand the basis of a prediction. This research addresses these challenges by developing a rice leaf disease classification system deployed on a mobile platform that is interpretable. The study utilizes the MobileNetV3-Large architecture, specifically chosen for its computational efficiency and high performance on edge devices. To provide model explainability, the Gradient-weighted Class Activation Mapping (Grad-CAM) method is integrated, which generates visual heatmaps highlighting the specific areas of the leaf that influenced the model's decision. Due to the technical limitations of mobile machine learning libraries regarding gradient computation, a client-server architecture was adopted. Disease classification is performed locally on the Android device using a TensorFlow Lite model for offline functionality, while Grad-CAM heatmaps are generated via an external server. The dataset used consists of images from Kaggle, covering four classes: Bacterial Leaf Blight, Blast, Brown Spot, and Healthy. Experimental results demonstrate that the MobileNetV3-Large attained high-performance metrics across various data split scenarios. The final integrated mobile application allows farmers to capture leaf images, receive immediate disease classifications, and view Grad-CAM heatmaps for visual validation. This approach provides a balance between high-accuracy mobile inference and meaningful model interpretability, supporting more effective and reliable rice cultivation practices.

Item Type:

Thesis (Undergraduate)

Contributors:

Contribution	Contributors	NIDN/NIDK	Email
Thesis advisor	Anggraeny, Fetty Tri	NIDN198202112021212005	fettyanggraeny.if@upnjatim.ac.id
Thesis advisor	Puspaningrum, Eva Yulia	NIDN198907052021212002	evapuspaningrum.if@upnjatim.ac.id

Subjects:

Q Science > QA Mathematics > QA76.87 Neural computers

Divisions:

Faculty of Computer Science > Departemen of Informatics

Depositing User:

Mr Adyatma Susanto

Date Deposited:

15 Jun 2026 06:24

Last Modified:

15 Jun 2026 06:38

URI:

https://repository.upnjatim.ac.id/id/eprint/53989