基于卷积神经网络的农作物病害检测研究综述  

A review of crop disease detection research based on convolutional neural networks

作　　者：乔世成[1] 党珊珊 何海祝 关强 王郝日钦 路扬 Qiao Shicheng;Dang Shanshan;He Haizhu;Guan Qiang;Wang Haoriqin;Lu Yang(College of Computer Science and Technology,Inner Mongolia Minzu University,Tongliao 028043,China)

机构地区：[1]内蒙古民族大学计算机科学与技术学院,内蒙古通辽028043

出　　处：《山西农业大学学报(自然科学版)》2025年第2期113-127,共15页Journal of Shanxi Agricultural University(Natural Science Edition)

基　　金：国家自然科学基金(62162049);内蒙古民族大学博士科研启动基金(BS658)。

摘　　要：我国是农业大国,拥有广大的农作物种植面积和丰富的农业资源。然而,近年来,农作物病害问题日益严重。农作物病害不仅直接影响产量和质量,还会造成农民的经济损失,威胁粮食安全和生态环境,对我国农业可持续发展构成了巨大威胁。因此,对农作物病害的精准检测是提高我国农业发展的关键因素。随着深度学习的不断发展,无损检测技术已得到广泛应用,利用卷积神经网络进行农作物病害的精准检测成为近年来研究的热点。卷积神经网络具有较好的图像检测与识别能力,能够适应多种病害类型,实现高效、准确的大规模检测,被广泛应用于农作物病害的精准检测中。本文首先介绍了卷积神经网络结构;然后探讨了几种典型的检测农作物病害的卷积神经网络模型;其次分析了其它神经网络研究情况并进行总结;重点讨论了目前基于小样本学习、小目标检测、网络轻量化改进的卷积神经网络热点研究问题;之后对未来农作物病害检测所面临的挑战和展望进行了总结,如针对数据集标注困难、模型缺乏泛化能力、小样本小目标数据集识别精度较低等问题,提出了建立更高质量的农作物病害数据集、优化小样本小目标数据集下的网络模型结构以及对农作物病害无损检测进行实时监测与预警等研究展望,以期为不断推进农业技术创新和应用、为我国农作物病害的精准检测研究提供参考依据。China is a major agricultural country with extensive crop cultivation areas and abundant agricultural resources.Howev-er,in recent years,crop disease problems have become increasingly severe,directly affecting yield and quality,causing eco-nomic losses for farmers,and posing threats to food security and the ecological environment.These issues have significantly hindered the sustainable development of agriculture in China.Therefore,precise detection of crop diseases is a key factor in en-hancing agricultural development in China.With the continuous development of deep learning,non-destructive detection tech-nologies have been widely applied.The use of convolutional neural networks(CNNs)for accurate crop disease detection has become a hot research topic in recent years.CNNs exhibit strong capabilities in image detection and recognition,can adapt to various types of diseases,and enable efficient and accurate large-scale detection,making them widely applicable in the precise detection of crop diseases.This paper first introduces the structure of convolutional neural networks,then explores several typi-cal convolutional neural network models for crop disease detection,and summarizes other neural network research.The paper focuses on current hot research topics in CNNs based on small-sample learning,small target detection,and network lightweight improvements.Additionally,it summarizes the challenges and future prospects for crop disease detection,including the difficul-ties in dataset annotation,lack of model generalization ability,and low recognition accuracy for small-sample and small-target datasets.It proposes research directions such as establishing higher-quality crop disease datasets,optimizing network model structures for small-sample and small-target datasets,and conducting real-time monitoring and early warning for non-destruc-tive crop disease detection.These insights aim to continuously promote agricultural technology innovation and application,pro-viding a reference for precise detection research of crop d

关 键 词：卷积神经网络 小样本 小目标 轻量化 

分 类 号：S126[农业科学—农业基础科学]