基于Deep-Semi-NMF的苹果斑点落叶病检测方法

doi:10.12133/j.smartag.SA202409001

摘要/Abstract

摘要：

［目的/意义］ 苹果斑点落叶病易导致苹果树叶过早脱落，从而影响苹果品质和产量。因此，如何准确检测此病一直是苹果树病害精准防治的热点问题。由于逆光等因素影响，传统基于图像分割的病斑检测方法难以在复杂背景下准确检测病斑区域边界，亟需发展苹果斑点落叶病检测新方法，助力苹果树病害精准防治。 ［方法］ 针对上述问题，本研究从图像异常检测的角度出发，考虑复杂背景干扰，采用深度半非负矩阵分解理论，结合鲁棒性好的马氏距离度量，提出一种新的深度半非负矩阵分解的马氏距离异常检测方法（Deep Semi-Non-Negative Matrix Factorization-Based Mahalanobis Distance-Anomaly Detector, DSNMFMAD）。该方法首先利用深度非负矩阵分解（Deep Semi-Non-Negative Matrix Factorization, DSNMF）提取图像中低秩的背景部分和稀疏的异常部分。然后采用基于奇异值分解特征子空间的马氏距离构建病斑检测器，检测器通过计算异常部分每个像元的异常度来标记病斑。最后，分别构建了实验室和自然条件下的两个苹果斑点落叶病数据集，用以验证提出方法的有效性。 ［结果和讨论］ DSNMFMAD在实验室条件和自然条件下对苹果斑点落叶病的识别准确率分别达到了99.8%和87.8%；平均检测速度为0.087和0.091 s/幅。相较于4种经典的异常检测方法和1种深度学习模型，本研究所提出方法的检测准确率在实验室条件下分别提高了0.2%、37.9%、10.3%、0.4%和24.5%；在自然条件下分别提高了2.5%、32.7%、5%、14.8%和3.5%。 ［结论］ 本研究提出的 DSNMFMAD能够通过DSNMF有效地将图像中的异常部分提取出来，并利用构建的病斑检测器准确地将苹果斑点落叶病位置检测出来。即使在复杂背景条件下，该方法亦获得了比对比方法更高的检测准确度，展现出了优异的病斑检测性能，为苹果斑点落叶病的检测与防治提供了技术参考依据。

关键词: 图像分割, 苹果斑点落叶病, 异常检测, 深度半非负矩阵分解, 马氏距离

Abstract:

[Objective] Apple Alternaria leaf spot can easily lead to premature defoliation of apple tree leaves, thereby affecting the quality and yield of apples. Consequently, accurately detecting of the disease has become a critical issue in the precise prevention and control of apple tree diseases. Due to factors such as backlighting, traditional image segmentation-based methods for detecting disease spots struggle to accurately identify the boundaries of diseased areas against complex backgrounds. There is an urgent need to develop new methods for detecting apple Alternaria leaf spot, which can assist in the precise prevention and control of apple tree diseases. [Methods] A novel detection method named Deep Semi-Non-negative Matrix Factorization-based Mahalanobis Distance Anomaly Detection (DSNMFMAD) was proposed, which combines Deep Semi-Non-negative Matrix Factorization (DSNMF) with Mahalanobis distance for robust anomaly detection in complex image backgrounds. The proposed method began by utilizing DSNMF to extract low-rank background components and sparse anomaly features from the apple Alternaria leaf spot images. This enabled effective separation of the background and anomalies, mitigating interference from complex background noise while preserving the non-negativity constraints inherent in the data. Subsequently, Mahalanobis distance was employed, based on the Singular Value Decomposition (SVD) feature subspace, to construct a lesion detector. The detector identified lesions by calculating the anomaly degree of each pixel in the anomalous regions. The apple tree leaf disease dataset used was provided by PaddlePaddle AI-Studio. Each image in the dataset has a resolution of 512×512 pixels, in RGB color format, and was in JPEG format. The dataset was captured in both laboratory and natural environments. Under laboratory conditions, 190 images of apple leaves with spot-induced leaf drop were used, while 237 images were collected under natural conditions. Furthermore, the dataset was augmented with geometric transformations and random changes in brightness, contrast, and hue, resulting in 1 145 images under laboratory conditions and 1 419 images under natural conditions. These images reflect various real-world scenarios, capturing apple leaves at different stages of maturity, in diverse lighting conditions, angles, and noise environments. This diversed dataset ensured that the proposed method could be tested under a wide range of practical conditions, providing a comprehensive evaluation of its effectiveness in detecting apple Alternaria leaf spot. [Results and Discussions] DSNMFMAD demonstrated outstanding performance under both laboratory and natural conditions. A comparative analysis was conducted with several other detection methods, including GRX (Reed-Xiaoli detector), LRX (Local Reed-Xiaoli detector), CRD (Collaborative-Representation-Based Detector), LSMAD (LRaSMD-Based Mahalanobis Distance Detector), and the deep learning model Unet. The results demonstrated that DSNMFMAD exhibited superior performance in the laboratory environment. The results demonstrated that DSNMFMAD attained a recognition accuracy of 99.8% and a detection speed of 0.087 2 s/image. The accuracy of DSNMFMAD was found to exceed that of GRX, LRX, CRD, LSMAD, and Unet by 0.2%, 37.9%, 10.3%, 0.4%, and 24.5%, respectively. Additionally, the DSNMFMAD exhibited a substantially superior detection speed in comparison to LRX, CRD, LSMAD, and Unet, with an improvement of 8.864, 107.185, 0.309, and 1.565 s, respectively. In a natural environment, where a dataset of 1 419 images of apple Alternaria leaf spot was analysed, DSNMFMAD demonstrated an 87.8% recognition accuracy, with an average detection speed of 0.091 0 s per image. In this case, its accuracy outperformed that of GRX, LRX, CRD, LSMAD, and Unet by 2.5%, 32.7%, 5%, 14.8%, and 3.5%, respectively. Furthermore, the detection speed was faster than that of LRX, CRD, LSMAD, and Unet by 2.898, 132.017, 0.224, and 1.825 s, respectively. [Conclusions] The DSNMFMAD proposed in this study was capable of effectively extracting anomalous parts of an image through DSNMF and accurately detecting the location of apple Alternaria leaf spot using a constructed lesion detector. This method achieved higher detection accuracy compared to the benchmark methods, even under complex background conditions, demonstrating excellent performance in lesion detection. This advancement could provide a valuable technical reference for the detection and prevention of apple Alternaria leaf spot.

Key words: image segment, apple Alternaria leaf spot, anomaly detection, deep semi-nonnegative matrix factorization, mahalanobis distance

中图分类号:

傅卓军, 胡政, 邓阳君, 龙陈锋, 朱幸辉. 基于Deep-Semi-NMF的苹果斑点落叶病检测方法[J]. 智慧农业(中英文), 2024, 6(6): 144-154.

FU Zhuojun, HU Zheng, DENG Yangjun, LONG Chenfeng, ZHU Xinghui. Detection Method of Apple Alternaria Leaf Spot Based on Deep-Semi-NMF[J]. Smart Agriculture, 2024, 6(6): 144-154.

图/表 16

图1

图2

图3

表1

表2

图4

表3

图 5

图6

表4

表5

图7

表6

图8

图9

表 7

参考文献 23

1	中国苹果产业协会, 国家苹果产业技术体系. 2023年度中国苹果产业发展报告[R]. 中国苹果产业协会, 2023.
2	CHAKRABORTY S, PAUL S, RAHAT-UZ-ZAMAN M. Prediction of apple leaf diseases using multiclass support vector machine[C]// 2021 2nd International Conference on Robotics, Electrical and Signal Processing Techniques (ICREST). Piscataway, New Jersey, USA: IEEE, 2021.
3	SINGH S, GUPTA S, TANTA A, et al. Extraction of multiple diseases in apple leaf using machine learning[J]. International journal of image and graphics, 2022, 22(3): ID 2140009.
4	KHAN A I, QUADRI S M K, BANDAY S, et al. Deep diagnosis: A real-time apple leaf disease detection system based on deep learning[J]. Computers and electronics in agriculture, 2022, 198: ID 107093.
5	AICH P, ATABY AAL, MAHYOUB M, et al. Automated plant disease diagnosis in apple trees based on supervised machine learning model[C]// 2023 15th International Conference on Developments in eSystems Engineering (DeSE). Piscataway, New Jersey, USA: IEEE, 2023.
6	AHMED I, YADAV P K. Predicting apple plant diseases in orchards using machine learning and deep learning algorithms[J]. SN computer science, 2024, 5(6): ID 700.
7	ZHAN B S, XIONG X, LI X L, et al. BHC-YOLOV8: Improved YOLOv8-based BHC target detection model for tea leaf disease and defect in real-world scenarios[J]. Frontiers in plant science, 2024, 15: ID 1492504.
8	WANG J L, QIN C C, HOU B B, et al. LCGSC-YOLO: A lightweight apple leaf diseases detection method based on LCNet and GSConv module under YOLO framework[J]. Frontiers in plant science, 2024, 15: ID 1398277.
9	李亚文, 何甜. 基于GA-SVM和特征提取的苹果叶部病害识别检测[J]. 食品与发酵科技, 2024, 60(3): 7-13.
	LI Y W, HE T. Detection of apple leaf disease identification based on GA-SVM and feature extraction[J]. Food and fermentation science & technology, 2024, 60(3): 7-13.
10	邵彧, 张善文, 李萍. 基于判别局部保持投影的苹果叶部病害识别方法[J]. 东北农业科学, 2021, 46(4): 113-118, 134.
	SHAO Y, ZHANG S W, LI P. Apple leaf disease recognition based on discriminant local preserving projection[J]. Journal of northeast agricultural sciences, 2021, 46(4): 113-118, 134.
11	ZHANG S L, WANG J Z, YANG K, et al. YOLO-ACT: An adaptive cross-layer integration method for apple leaf disease detection[J]. Frontiers in plant science, 2024, 15: ID 1451078.
12	WU Z, SU H, TAO X, et al. Adaptive dictionary construction for hyperspectral anomaly detection based on collaborative representation[C]// 2022 IEEE International Geoscience and Remote Sensing Symposium. Piscataway, New Jersey, USA: IEEE, 2022.
13	AI S. Pathological image of apple leaf [DS/OL]. (2021-08-18)[2023-0610].
14	SHYAMALA DEVI M, ESWAR R, M D H R, et al. Encrypt decrypt ReLU activated UNet prototype based prediction of leaf disease segmentation[C]// 2024 International Conference on Signal Processing, Computation, Electronics, Power and Telecommunication (IConSCEPT). Piscataway, New Jersey, USA: IEEE, 2024.
15	黄成龙, 张忠福, 卢智浩, 等. 基于VFNet-Improved和DeepSort的棉花黄萎病病情分级[J]. 智能化农业装备学报(中英文), 2023, 4 (2): 12-21.
	HUANG C L, ZHANG Z F, LU Z H, et al. Leaf gradingfor cotton verticillium wilt based on VFNet-Improved and DeepSort[J]. Journal of intelligent agricultural mechanization, 2023(2): 12-21.
16	CHEN W S, ZENG Q W, PAN B B. A survey of deep nonnegative matrix factorization[J]. Neurocomputing, 2022, 491: 305-320.
17	DE HANDSCHUTTER P, GILLIS N, SIEBERT X. A survey on deep matrix factorizations[J]. Computer science review, 2021, 42: ID 100423.
18	TRIGEORGIS G, BOUSMALIS K, ZAFEIRIOU S, et al. A deep matrix factorization method for learning attribute representations[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(3): 417-429.
19	REED I S, YU X L. Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distribution[J]. IEEE transactions on acoustics speech and signal processing, 1990, 38: 1760-1770.
20	MOLERO J M, GARZÓN E M, GARCÍA I . et al. Analysis and optimizations of global and local versions of the RX algorithm for anomaly detection in hyperspectral data[J]. IEEE journal of selected topics in applied earth observations and remote sensing, 2013, 6(2): 801-814.
21	LI W, DU Q. Collaborative representation for hyperspectral anomaly detection[J]. IEEE transactions on geoscience and remote sensing, 2015, 53(3): 1463-1474.
22	ZHANG Y X, DU B, ZHANG L P, et al. A low-rank and sparse matrix decomposition-based mahalanobis distance method for hyperspectral anomaly detection[J]. IEEE transactions on geoscience and remote sensing, 2016, 54(3): 1376-1389.
23	唐辉, 王铭, 于秋实, 等. 融合改进UNet和迁移学习的棉花根系图像分割方法[J]. 智慧农业(中英文), 2023, 5 (3): 96-109.
	TANG H, WANG M, YU Q S, et al. Root image segmentation method based on improved UNet and transfer learning[J]. Smart agriculture, 2023, 5(3): 96-109.

试验环境	配置参数
操作系统	Windows10
处理器	Core（TM）i9-12900K
内存容量	128 G
编程语言	MATLAB2022b
显卡	NVIDIA RTX 4090 24 G

Layer	ROC曲线面积	平均运行时间/s
1	0.997 500	0.073 4
2	0.993 070	0.087 2
3	0.874 530	0.110 4

数据类型	数量/张	方法	GRX	LRX	CRD	LSMAD	Unet	DSNMFMAD
Indoor	1 145	Average_AUC	0.995 5	0.616 6	0.893 2	0.993 9	0.752 2	0.997 5
Indoor	1 145	Average_Time/s	0.071 1	8.951 1	107.271 9	0.396 2	1.652 2	0.087 2

Layer	ROC曲线面积	平均运行时间/s
1	0.857 280	0.073 4
2	0.877 765	0.091 0
3	0.725 340	0.110 4

数据类型	数量/张	方法	GRX	LRX	CRD	LSMAD	Unet	DSNMFMAD
Outdoor	1 419	Average_AUC	0.852 518	0.551 311	0.828 772	0.729 894	0.842 670	0.877 765
Outdoor	1 419	Average_Time/s	0.070 6	2.987 8	132.108 3	0.315 0	1.91 60	0.091 0