Skip to content
SDGs Center UDINUS

Evidence record · UDINUS · SDGs

A Transfer Learning-Based Hybrid Deep Learning Framework for Multi-Crop Plant Leaf Disease Classification

Rektorat UDINUS 2026 Published 2026-10-30

Evidence record · UDINUS · SDGs

Unit
Rektorat UDINUS
Period
2026
Goals
Record
d35b8a1a-e93b-4389-a363-3ad5bb3085a4
Published
2026-10-30

Universitas Dian Nuswantoro, through the participation of Anton, Supriadi Rustad, Guruh Fajar Shidik, and Abdul Syukur, conducted research on a transfer learning-based hybrid deep learning framework for multi-crop plant leaf disease classification. The activity addressed the importance of early and accurate detection of plant leaf diseases because such diseases threaten global agricultural productivity and food security. In relation to the stated goals, the research focused on plant disease classification as an activity connected in the source to sustainable crop management and food security.

The research developed a multi-class classification framework for plant leaf diseases in cassava, wheat, and tomato. The architecture integrated DenseNet121 and a scaled MobileNetV2 with an alpha value of 0.35 to balance representation power and computational efficiency. The framework incorporated attention mechanisms, compressed feature maps using 1 × 1 convolutions, and fused the features in a compact classification head using Swish activation and batch normalization. Grad-CAM++ was used for model interpretability by highlighting disease-relevant regions.

The framework was evaluated on three public datasets comprising 10,635 images across 13 disease classes and one healthy class. The reported results included perfect accuracy on cassava, approximately 95% accuracy on wheat, and up to 99.6% accuracy on tomato. Despite a compact design with approximately 10 million parameters, the model performed competitively according to the source. Its inference latency was under 60 milliseconds, with throughput above 20 Frame per Second. The source also identified the framework as suitable for deployment on edge devices.

Through this activity, UDINUS contributed research addressing plant leaf disease detection, computational efficiency, and interpretability within multi-crop classification. The activity is institutionally framed as research relevant to sustainable crop management and food security, with its scope limited to the methods, datasets, classes, and findings explicitly reported in the source.

Reference https://www.scopus.com/pages/publications/105039130604?origin=resultslist