From 913a6145e83d98298bc9d4f015c0c7a3d68df706 Mon Sep 17 00:00:00 2001 From: GunjanBasak007 Date: Wed, 11 Feb 2026 13:54:46 +0530 Subject: [PATCH] Add Responsibility document for disease classification study This document outlines a lightweight, attention-enhanced hybrid approach for classifying maize leaf diseases using MobileNetV2 and SVM, detailing methodology and results. --- Responsibility | 14 ++++++++++++++ 1 file changed, 14 insertions(+) create mode 100644 Responsibility diff --git a/Responsibility b/Responsibility new file mode 100644 index 0000000..1c925cc --- /dev/null +++ b/Responsibility @@ -0,0 +1,14 @@ +A Lightweight Attention-Enhanced CNN-SVM Hybrid Approach for Maize Leaf Disease Classification: + +MD Soyeb Hoque1, Gunjan Basak1, Dipan Deb1, Nikhil Sah1, Arghyasree Saha1 [0009-0006-5879-6445] and Pawan Kumar Singh1,* [0000-0002-9598-7981] +1 Department of Information Technology, Jadavpur University, Salt Lake Campus, Kolkata-700106, India + +{hoquesoyeb9@gmail.com, gunjanbasak007@gmail.com, deb.dipan2004@gmail.com, nikhil.jadavpur@gmail.com, arghyasree07@gmail.com, pawansingh.ju@gmail.com} +*Corresponding author: Pawan Kumar Singh {pawansingh.ju@gmail.com} + +Maize productivity is threatened by leaf diseases such as curvularia leaf spot, small spot, and rust, making early detection essential. Manual diagnosis is labor-intensive and error-prone, +while many deep learning (DL) models are computationally demanding for practical deployment. This study introduces a lightweight, attention-enhanced hybrid framework that combines +MobileNetV2 with a Convolutional Block Attention Module (CBAM) for discriminative feature extraction along with a Support Vector Machine (SVM) for robust disease classification. +Extensive experiments on two benchmark datasets demonstrate classification accuracies of 98.11% and 99.44%. In addition, a comprehensive ablation study is performed to analyze +the impact of attention mechanism and hybrid learning, confirming their effectiveness in improving performance and generalization. Overall, the results highlight the proposed framework’s +efficiency and suitability for real-world agricultural applications.