Related Experiment Video
Updated: Oct 16, 2025
08:20
Author Spotlight: AI-Driven Trypanosome Species Detection from Microscopic Images
Published on: October 27, 2023
1.8K
Diagnosis of Typical Apple Diseases: A Deep Learning Method Based on Multi-Scale Dense Classification Network.
Yunong Tian1,2, En Li1,2, Zize Liang1,2
1State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing, China.
Frontiers in Plant Science
|October 18, 2021
Summary
Accurate apple disease diagnosis is crucial for agriculture. A new Multi-scale Dense classification network, enhanced by Cycle-GAN for data augmentation, achieved state-of-the-art accuracy in identifying 11 types of apple diseases.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- Apple quality and yield decline due to diseases significantly impact the agricultural economy.
- Precise diagnosis and decision-making are vital for reducing losses and promoting economic growth.
Purpose of the Study:
- To develop a novel Multi-scale Dense classification network for diagnosing 11 types of apple diseases (fruits and leaves).
- To improve diagnostic performance by addressing insufficient image data for specific diseases like anthracnose and ring rot.
Main Methods:
- Dataset expansion using the Cycle-GAN algorithm to generate realistic disease lesions.
- Development of two models: Multi-scale Dense Inception-V4 and Multi-scale Dense Inception-Resnet-V2, utilizing DenseNet and multi-scale connections.
- Training and evaluation of models on 11 image types, including healthy and diseased apples with varying disease severities.
Main Results:
- Cycle-GAN effectively expanded the dataset, improving diagnostic performance over traditional augmentation.
- The proposed Multi-scale Dense Inception-V4 and Multi-scale Dense Inception-Resnet-V2 models achieved classification accuracies of 94.31% and 94.74%, respectively.
- Both models outperformed the DenseNet-121 network, reaching state-of-the-art performance.
Conclusions:
- The novel Multi-scale Dense classification network, combined with Cycle-GAN data augmentation, offers a highly effective solution for apple disease diagnosis.
- This approach significantly enhances diagnostic accuracy, contributing to reduced agricultural losses and economic growth.
