Skip to content

AbhinandanMandal/MoSE-XAI

Repository files navigation

MoSE-XAI (Accepted COMSNETS Graduate Forum 2026)

ChatGPT Image Jun 8, 2026, 12_46_50 PM

This project implements a modified MobileNetV2 architecture enhanced with a Squeeze-and-Excitation (SE) block to improve feature recalibration and model performance in crop disease detection for soybean and sugarcane. To ensure transparency and interpretability, the model integrates Explainable AI (XAI) techniques using Grad-CAM for visual explanation of predictions.


Features

  • MobileNetV2 Architecture Lightweight and efficient convolutional neural network designed for mobile and embedded vision applications.

  • Squeeze-and-Excitation (SE) Block Enhances channel-wise attention by adaptively recalibrating feature responses.

  • Explainable AI (Grad-CAM) Generates heatmaps to visualize which regions of the input image influence the model’s decision.


Model Architecture Overview

  • Backbone model: MobileNetV2
  • Added module: Squeeze-and-Excitation (SE) block
  • XAI method: Grad-CAM
  • Goal: Crop disease detection with visual explainability

Dataset

To train MoSE model, the author used 3 different datasets

  • sugarcane leaf image dataset
  • sugarcane leaf disease dataset
  • soynet: indian soybean image dataset

All datas are collected from Mendeley Data.


Thanks

Abhinadan Mandal

About

MoSE-XAI: a lightweight, interpretable, deployable model for crop disease detection

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors