基于深度学习的苹果叶部主要病害分类方法

Classification methods of main apple leaf diseases based on deep learning

摘要

自然场景下叶部病害识别面临自动化与精准化难题，基于深度学习构建识别模型是解决叶部病害精准识别难题的新途径。针对自然场景下叶部病害背景复杂、识别准确率低、终端设备算力受限等问题，本研究构建了叶部病害分类图像数据集，并通过数据增强、高斯滤波去噪、标准化等方式，对图像数据集进行了预处理；采用传统机器学习、卷积神经网络及迁移学习轻量化网络模型，分别构建叶部病害分类模型。对比不同模型发现，以卷积神经网络为基础，引入 SqueezeNet 所构建的“卷积神经网络+数据增强+迁移学习+优化改进”轻量化网络模型，识别准确率达到 94.50%，较卷积神经网络提升约 13.00%。通过混淆矩阵及受试者工作特征曲线(receive operating characteristic，ROC)评估验证，发现所构建模型对 6 种叶部病害分类准确率均超 85.00%，受试者工作特征曲线下面积(area under the curve，AUC)均超 0.92，表明该模型能够实现叶部病害精准分类，为苹果叶部病害的快速识别提供技术支撑，对果园病害早发现、早防治具有实际参考价值。

Abstract

Automated and accurate identification of leaf diseases in natural scenes faces prominent challenges，and constructing recognition models based on deep learning provides a novel approach to address the precise leaf disease identification. To solve the problems of complex backgrounds，low recognition accuracy，and limited computing power of terminal devices in leaf disease recognition under natural scenarios，this study constructs an image dataset for leaf disease classification，and preprocesses the dataset through data augmentation，Gaussian filtering denoising，and standardization. Three categories of leaf disease classification models are established，including traditional machine learning models，convolutional neural network (CNN) models，and lightweight network models based on transfer learning. The comparison of different models shows that the lightweight network model，which is built on CNN and integrated with SqueezeNet under the framework of "CNN + data augmentation + transfer learning + optimization improvement"，achieves a recognition accuracy of 94.50%，which is approximately 13.00% higher than that of the basic CNN model. Verified by the confusion matrix and Receiver Operating Characteristic (ROC) curve evaluation，the proposed model yields a classification accuracy of over 85.00% for six types of leaf diseases，with all Area Under the Curve (AUC) values exceeding 0.92，indicating that the model can realize accurate classification of leaf diseases. This study provides technical support for the rapid identification of apple leaf diseases，and has practical reference value for the early detection and prevention of orchard diseases.

关键词

叶部病害识别 / 卷积神经网络 / SqueezeNet / 迁移学习 / 轻量化模型

Key words

leaf disease identification / CNN / SqueezeNet / transfer learning / lightweight model

参考文献

[1] 陈旭, 解天祺. 2020. 基于 B—CLBP 和 GLCM 特征的纹理材质分类[J]. 计算机应用与软件, 37(6):242-246.

[2] Chen X, Xie T Q . 2020. Texture material classification based on b—clbp and glcm features[J]. Computer Applications and Software, 37(6):242-246.(in Chinese)

[3] 方莉, 张萍. 2010. 经典图像去噪算法研究综述[J]. 工业控制计算机, 23(11):73-74.

[4] Fang L, Zhang P . 2010. Overveiw on some arithmetics for image denoising[J]. Industrial Control Computer, 23(11):73-74.(in Chinese)

[5] 李想, 胡肖楠, 李方一, 等. 2023. 苹果树叶多病害及不可辨别病害的轻量识别算法[J]. 农业工程学报, 39(14):184-190.

[6] Li X, Hu X N, Li F Y, et al. 2023. Lightweight recognition for multiple and indistinguishable diseases of apple tree leaf[J]. Transactions of the Chinese Society of Agricultural Engineering, 39(14):184-190.(in Chinese)

[7] 李宗儒, 何东健. 2010. 基于手机拍摄图像分析的苹果病害识别技术研究[J]. 计算机工程与设计, 31(13):3051−3053， 3095.

[8] Li Z R, He D J . 2010. Research on identify technologies of apple’s disease based on mobile photograph image analysis[J]. Computer Engineering and Design, 31(13):3051−3053， 3095.(in Chinese)

[9] 刘合兵, 鲁笛, 席磊. 2022. 基于 MobileNetV2 和迁移学习的玉米病害识别研究[J]. 河南农业大学学报, 56(6):1041-1051.

[10] Liu H B, Lu D, Xi L . 2022. The research of maize disease identification based on MobileNetV2 and transfer learning[J]. Journal of Henan Agricultural University, 56(6):1041-1051.(in Chinese)

[11] 满凤环, 陈秀宏, 何佳佳. 2018. 改进的 Dropout 正则化卷积神经网络[J]. 传感器与微系统, 37(4):44-47.

[12] Man F H, Chen X H, He J J . 2018. Improved dropout regularization CNN[J]. Transducer and Microsystem Technologies, 37(4):44-47.(in Chinese)

[13] 明浩, 苏喜友. 2020. 利用特征分割和病斑增强的杨树叶部病害识别[J]. 浙江农林大学学报, 37(6):1159-1166.

[14] Ming H, Su X Y . 2020. Image recognition of poplar leaf diseases with feature segmentation and lesion enhancement[J]. Journal of Zhejiang A & F University, 37(6):1159-1166.(in Chinese)

[15] 孙俊, 谭文军, 毛罕平, 等. 2017. 基于改进卷积神经网络的多种植物叶片病害识别[J]. 农业工程学报, 33(19):209-215.

[16] Sun J, Tan W J, Mao H P, et al. 2017. Recognition of multiple plant leaf diseases based on improved convolutional neural network[J]. Transactions of the Chinese Society of Agricultural Engineering, 33(19):209-215.(in Chinese)

[17] 孙文杰, 牟少敏, 董萌萍, 等. 2020. 基于卷积循环神经网络的桃树叶部病害图像识别[J]. 山东农业大学学报(自然科学版), 51(6):998-1003.

[18] Sun W J, Mu S M, Dong M P, et al. 2020. Image recognition of peach leaf diseases based on convolutional recurrent neural network[J]. Journal of Shandong Agricultural University (Natural Science Edition), 51(6):998-1003.(in Chinese)

[19] 田呈明, 张星耀. 2024. 树木医学的机遇与挑战[J]. 树木医学, 1( 1):1-8.

[20] Tian C M, Zhang X Y . 2024. Opportunities and challenges in tree medicine[J]. Tree Health, 1( 1):1-8.(in Chinese)

[21] 田有文, 李天来, 李成华, 等. 2007. 基于支持向量机的葡萄病害图像识别方法[J]. 农业工程学报, 23(6):175-180.

[22] Tian Y W, Li T L, Li C H, et al. 2007. Method for recognition of grape disease based on support vector machine[J]. Transactions of the Chinese Society of Agricultural Engineering, 23(6):175-180.(in Chinese)

[23] 王哲豪, 范丽丽, 何前. 2023. 基于 MobileNet V2 和迁移学习的番茄病害识别[J]. 江苏农业科学, 51(9):215-221.

[24] Wang Z H, Fan L L, He Q . 2023. Recognition of tomato disease based on transfer learning and MobileNet V2[J]. Jiangsu Agricultural Sciences, 51(9):215-221.(in Chinese)

[25] 许景辉, 邵明烨, 王一琛, 等. 2020. 基于迁移学习的卷积神经网络玉米病害图像识别[J]. 农业机械学报, 51(2):230−236， 253.

[26] Xu J H, Shao M Y, Wang Y C, et al. 2020. Recognition of corn leaf spot and rust based on transfer learning with convolutional neural network[J]. Transactions of the Chinese Society for Agricultural Machinery, 51(2):230−236， 253.(in Chinese)

[27] 严理, 夏承博, 温远光. 2016. 土壤原生病原体对森林植被及生态系统的影响[J]. 世界林业研究, 29(5):22-28.

[28] Yan L, Xia C B, Wen Y G . 2016. The influence of soil—borne pathogens on forest vegetation and ecosystem[J]. World Forestry Research, 29(5):22-28.(in Chinese)

[29] 杨方. 2020. 基于深度学习的梨树叶部病害识别研究[D]. 晋中: 山西农业大学.

[30] Yang F. 2020. Research on pear leaf disease recognition based on deep learning[D]. Taigu: Shanxi Agricultural University.(in Chinese)

[31] 赵立新, 侯发东, 吕正超, 等. 2020. 基于迁移学习的棉花叶部病虫害图像识别[J]. 农业工程学报, 36(7):184-191.

[32] Zhao L X, Hou F D, Lyu Z C, et al. 2020. Image recognition of cotton leaf diseases and pests based on transfer learning[J]. Transactions of the Chinese Society of Agricultural Engineering, 36(7):184-191.(in Chinese)

[33] 周志华. 2016. 机器学习[M]. 北京: 清华大学出版社: 35‒ 37.

[34] Zhou Z H . 2016. Machine Learning[M]. Beijing: Tsinghua University Press: 35‒ 37.(in Chinese)

[35] Chen J D, Zhang D F, Nanehkaran Y A. 2020. Identifying plant diseases using deep transfer learning and enhanced lightweight network[J]. Multimedia Tools and Applications, 79(41/42):31497-31515.

[36] Kaur P, Harnal S, Gautam V ,et al. 2023. A novel transfer deep learning method for detection and classification of plant leaf disease[J]. Journal of Ambient Intelligence and Humanized Computing, 14(9):12407-12424.

[37] Liu J, Wang X W. 2020. Early recognition of tomato gray leaf spot disease based on MobileNetv2—YOLOv3 model[J]. Plant Methods, 16: 83.

[38] Mohanty S P, Hughes D P, Salathé M. 2016. Using deep learning for image—based plant disease detection[J]. Frontiers in Plant Science, 7: 1419.

[39] Pietikäinen M. 2010. Local binary patterns[J]. Scholarpedia, 5(3):9775.

[40] Woo S, Debnath S, Hu R H ,et al. 2023. ConvNeXt V2:Co—designing and scaling ConvNets with masked autoencoders[C]. Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR): 16133‒ 16142.

[41] Yang Q, Duan S K, Wang L D. 2022. Efficient identification of apple leaf diseases in the wild using convolutional neural networks[J]. Agronomy, 12(11):2784.

[42] Zeng W H, Li M. 2020. Crop leaf disease recognition based on Self‒Attention convolutional neural network[J]. Computers and Electronics in Agriculture, 172: 105341.

基金资助

国家自然科学基金项目(31870537)

黑龙江省自然科学基金项目(LH2024C054)