华东师范大学学报（自然科学版） ›› 2025, Vol. 2025 ›› Issue (1): 59-71.doi: 10.3969/j.issn.1000-5641.2025.01.005

• 计算机科学 • 上一篇    下一篇

基于知识蒸馏的轻量化农作物病害识别算法

胡雯婧1, 蒋龙泉2, 余俊龙1, 徐伊茜3,*(), 刘奇鹏4, 梁雷5, 李嘉豪4   

  1. 1. 福州大学 梅努斯国际工程学院, 福州　350108
    2. 工业互联网创新中心(上海)有限公司, 上海　200131
    3. 复旦大学 工程与应用技术研究院, 上海　200433
    4. 上海兰桂骐技术发展股份有限公司, 上海　200131
    5. 牧星智能工业科技(上海)有限公司, 上海　200131
  • 收稿日期:2023-12-26 出版日期:2025-01-25 发布日期:2025-01-20
  • 通讯作者: 徐伊茜 E-mail:xuyiqian@fudan.edu.cn
  • 基金资助:
    临港新片区高新产业和科技创新专项项目 (SH-LG-GK-2020-02-11); 类脑智能科技有限公司和上海类脑芯片与片上智能系统研发与转化功能型平台项目 (17DZ2260900)

Knowledge-distillation-based lightweight crop-disease-recognition algorithm

Wenjing HU1, Longquan JIANG2, Junlong YU1, Yiqian XU3,*(), Qipeng LIU4, Lei LIANG5, Jiahao LI4   

  1. 1. Maynooth International Engineering College, Fuzhou University, Fuzhou　350108, China
    2. Industrial Internet Innovation Center (Shanghai) Co. Ltd., Shanghai　200131, China
    3. Academy for Engineering & Technology, Fudan University, Shanghai　200433, China
    4. Shanghai Lankaukei Technology Development Co. Ltd., Shanghai　200131, China
    5. Muxing Intelligent Industrial Technology (Shanghai) Co. Ltd., Shanghai　200131, China
  • Received:2023-12-26 Online:2025-01-25 Published:2025-01-20
  • Contact: Yiqian XU E-mail:xuyiqian@fudan.edu.cn

RichHTML

17

PDF

374

摘要：

农作物病害是威胁农作物生长的主要因素之一, 机器学习算法能高效率实现大范围农作物病害的发现, 有利于对其进行及时处理, 进而提升农作物的产量和质量. 在大范围农业场景中, 由于供电等条件限制, 无法满足服务器等高算力设备的供电需求, 现有深度网络模型大多需要较高算力, 难以部署在低功耗的嵌入式设备上, 给大范围农作物病害的准确识别应用带来障碍. 为解决此问题, 提出了一种基于知识蒸馏的轻量化农作物病害识别模型, 并设计了一种基于残差结构和注意力机制的学生模型, 利用知识蒸馏方法从大规模模型ConvNeXt中迁移学习成果, 在实现模型轻量化的同时保持高精度识别. 实验结果表明, 在模型规模为2.28 MB的条件下, 39类农作物病害图像分类任务的准确率达到了98.72%, 且每类病害的精确率、召回率和特异度均高于90%. 该模型满足了在嵌入式设备中部署的需求, 为农作物病害识别提供了一种实用高效的解决方法.

关键词: 农作物病害识别, 卷积神经网络, 知识蒸馏算法, 注意力机制

Abstract:

Crop diseases are one of the main factors threatening crop growth. In this regard, machine-learning algorithms can efficiently detect large-scale crop diseases and are beneficial for timely processing and improving crop yield and quality. In large-scale agricultural scenarios, owing to limitations in power supply and other conditions, the power-supply requirements of high-computing-power devices such as servers cannot be fulfilled. Most existing deep-network models require high computing power and cannot be deployed easily on low-power embedded devices, thus hindering the accurate identification and application of large-scale crop diseases. Hence, this paper proposes a lightweight crop-disease-recognition algorithm based on knowledge distillation. A student model based on a residual structure and the attention mechanism is designed and knowledge distillation is applied to complete transfer learning from the ConvNeXt model, thus achieving the lightweight model while maintaining high-precision recognition. The experimental results show that the accuracy of image classification for 39 types of crop diseases is 98.72% under a model size of 2.28 MB, which satisfies the requirement for deployment in embedded devices and indicates a practical and efficient solution for crop-disease recognition.

Key words: crop disease identification, convolutional neural networks, knowledge distillation algorithms, attention mechanisms

中图分类号: