Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LWMSCNN-SE: A Lightweight Multi-Scale Network for Efficient Maize Disease Classification on Edge Devices

Fikadu Weloday, Jianmei Su

TL;DR

LWMSCNN-SE tackles maize disease classification on resource-constrained edge devices by introducing a lightweight multi-scale CNN that combines residual depthwise blocks with Squeeze-and-Excitation attention. It achieves high accuracy (96.63%) with a compact footprint of 241,348 parameters and 0.666 GFLOPs on a four-class maize dataset, enabling real-time inference on mobile and embedded platforms. The method centers on Residual Multi-Scale Blocks for multi-scale feature extraction and SE-based channel recalibration, with comprehensive ablation and comparative analyses confirming favorable accuracy-efficiency trade-offs. The work also outlines training, augmentation, and deployment considerations, plus avenues for hardware-aware optimizations and broader dataset evaluation.

Abstract

Maize disease classification plays a vital role in mitigating yield losses and ensuring food security. However, the deployment of traditional disease detection models in resource-constrained environments, such as those using smartphones and drones, faces challenges due to high computational costs. To address these challenges, we propose LWMSCNN-SE, a lightweight convolutional neural network (CNN) that integrates multi-scale feature extraction, depthwise separable convolutions, and squeeze-and-Excitation (SE) attention mechanisms. This novel combination enables the model to achieve 96.63% classification accuracy with only 241,348 parameters and 0.666 GFLOPs, making it suitable for real-time deployment in field applications. Our approach addresses the accuracy--efficiency trade-off by delivering high accuracy while maintaining low computational costs, demonstrating its potential for efficient maize disease diagnosis on edge devices in precision farming systems.

LWMSCNN-SE: A Lightweight Multi-Scale Network for Efficient Maize Disease Classification on Edge Devices

TL;DR

LWMSCNN-SE tackles maize disease classification on resource-constrained edge devices by introducing a lightweight multi-scale CNN that combines residual depthwise blocks with Squeeze-and-Excitation attention. It achieves high accuracy (96.63%) with a compact footprint of 241,348 parameters and 0.666 GFLOPs on a four-class maize dataset, enabling real-time inference on mobile and embedded platforms. The method centers on Residual Multi-Scale Blocks for multi-scale feature extraction and SE-based channel recalibration, with comprehensive ablation and comparative analyses confirming favorable accuracy-efficiency trade-offs. The work also outlines training, augmentation, and deployment considerations, plus avenues for hardware-aware optimizations and broader dataset evaluation.

Abstract

Maize disease classification plays a vital role in mitigating yield losses and ensuring food security. However, the deployment of traditional disease detection models in resource-constrained environments, such as those using smartphones and drones, faces challenges due to high computational costs. To address these challenges, we propose LWMSCNN-SE, a lightweight convolutional neural network (CNN) that integrates multi-scale feature extraction, depthwise separable convolutions, and squeeze-and-Excitation (SE) attention mechanisms. This novel combination enables the model to achieve 96.63% classification accuracy with only 241,348 parameters and 0.666 GFLOPs, making it suitable for real-time deployment in field applications. Our approach addresses the accuracy--efficiency trade-off by delivering high accuracy while maintaining low computational costs, demonstrating its potential for efficient maize disease diagnosis on edge devices in precision farming systems.
Paper Structure (16 sections, 1 equation, 5 figures, 4 tables)

This paper contains 16 sections, 1 equation, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Methodology
  4. Data Description and Preprocessing
  5. Data Augmentation
  6. Overall Architecture
  7. Residual Multi-Scale Block
  8. Squeeze-and-Excitation Attention
  9. Network Architecture
  10. Experimental Results
  11. Training Configuration
  12. Classification Report and Confusion Matrix
  13. Training Performance Analysis and Visualization
  14. Comparison with Existing Models
  15. Ablation Study
  16. ...and 1 more sections

Figures (5)

  • Figure 1: End-to-end workflow of the proposed LWMSCNN-SE model for maize disease classification.
  • Figure 2: The architecture of the proposed LWMSCNN-SE model.
  • Figure 3: The architecture of the Residual Multi-Scale Block (RMSB).
  • Figure 4: Confusion Matrix for Maize Disease Classification.
  • Figure 5: Model training performance: Accuracy and loss curves.