Enriched Proximal Gradient Descent Deep Unfolding Framework for Rice Leaf Disease Classification

Rice leaf diseases create a critical threat to global food security, which requires rapid and effective diagnostic solutions.Existing Deep Learning (DL) frameworks, such as Residual Network (ResNet) and Visual Geometry Group Network (VGGNet) struggle to handle variations in lighting, background canopy clutter, and visual metamerism, which reduces the detection performance under various environmental conditions.To address these challenges, this research introduces the Enriched Proximal Gradient Descent Deep Unfolding Framework (EPGD-DUF) to develop an environmentally robust model to reduce sensitivity towards domain-specific illumination and background variations for rice-leaf disease detection.EPGD-DUF architecture employs the Proximal Gradient Descent (PGD) algorithm across four parameter-shared stages to mathematically model software-synthesized pseudo-spectral latent reconstruction.This integrates a Cross-Channel Latent Fusion Module (CCLFM) to suppress soil-line vectors and applies a Hyperspherical ArcFace classification head to strictly separate angular classes.The detection performance of the EPGD-DUF is evaluated on two standard datasets, achieving 99.83% accuracy on the laboratorycontrolled Mendeley rice leaf disease dataset and 98.89% accuracy on the highly cluttered Paddy Doctor field dataset.The proposed EPGD-DUF achieves higher accuracy than other baseline models, such as ResNet50, VGG16, ConvNeXt, and Vanilla PINN baselines under the corresponding dataset specific evaluation protocols.Under an explicit laboratory-to-field transfer protocol over the four classes common to both repositories, the framework retains 96.2% accuracy on the held-out target-domain test set, while target-domain labels were not used during adaptation or model selection, corresponding to a domain gap of 3.6 percentage points, compared with 6.0 percentage points for the strongest convolutional baseline.

Authors

Institutions

Publication Details

Journal
International journal of intelligent engineering and systems
Published
2026-11-17
DOI
https://doi.org/10.22266/ijies2026.1231.09
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Enriched Proximal Gradient Descent Deep Unfolding Framework for Rice Leaf Disease Classification

Rakesh Kancharla, Prasadarao Jala, Juvvala Bala Ambedkar, Uday Tummalacheruvu et al.
1 links
International journal of intelligent engineering and systems
Smart Agriculture and AI
article

Enriched Proximal Gradient Descent Deep Unfolding Framework for Rice Leaf Disease Classification

Rakesh Kancharla, Prasadarao Jala, Juvvala Bala Ambedkar, Uday Tummalacheruvu, Prasanth Kumar Godi
article en

Abstract

Rice leaf diseases create a critical threat to global food security, which requires rapid and effective diagnostic solutions.Existing Deep Learning (DL) frameworks, such as Residual Network (ResNet) and Visual Geometry Group Network (VGGNet) struggle to handle variations in lighting, background canopy clutter, and visual metamerism, which reduces the detection performance under various environmental conditions.To address these challenges, this research introduces the Enriched Proximal Gradient Descent Deep Unfolding Framework (EPGD-DUF) to develop an environmentally robust model to reduce sensitivity towards domain-specific illumination and background variations for rice-leaf disease detection.EPGD-DUF architecture employs the Proximal Gradient Descent (PGD) algorithm across four parameter-shared stages to mathematically model software-synthesized pseudo-spectral latent reconstruction.This integrates a Cross-Channel Latent Fusion Module (CCLFM) to suppress soil-line vectors and applies a Hyperspherical ArcFace classification head to strictly separate angular classes.The detection performance of the EPGD-DUF is evaluated on two standard datasets, achieving 99.83% accuracy on the laboratorycontrolled Mendeley rice leaf disease dataset and 98.89% accuracy on the highly cluttered Paddy Doctor field dataset.The proposed EPGD-DUF achieves higher accuracy than other baseline models, such as ResNet50, VGG16, ConvNeXt, and Vanilla PINN baselines under the corresponding dataset specific evaluation protocols.Under an explicit laboratory-to-field transfer protocol over the four classes common to both repositories, the framework retains 96.2% accuracy on the held-out target-domain test set, while target-domain labels were not used during adaptation or model selection, corresponding to a domain gap of 3.6 percentage points, compared with 6.0 percentage points for the strongest convolutional baseline.

International journal of intelligent engineering and systemsVol. 19(12)
Aditya University (IN), Koneru Lakshmaiah Education Foundation (IN)
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon