Original papersLeaf image based cucumber disease recognition using sparse representation classification
Introduction
Crop diseases have critical effects on the quality and quantity of agricultural products by destroying the normal state of the crop and change or interrupt crop vital functions such as photosynthesis, transpiration, pollination, fertilization, germination (Wang et al., 2015). Accurate recognition and diagnosis of crop diseases at early stage is very important (Gavhale and Ujwalla, 2014, Rumpf et al., 2010). The symptoms of crop diseases often appear visually on plant leaves, thus it is possible to automatically detect crop diseases by applying machine learning techniques on leaf images (Camargo and Smith, 2009, Hillnhuetter and Mahlein, 2008). Specifically, crop diseases can be classified and recognized by analyzing color, texture, and shape of the diseased leaf images (Al-Hiary et al., 2011, Gulhane and Gurjar, 2011, Patil and Kumar, 2011). Using features extracted from leaf images, many crop disease recognition methods and systems have been developed (Chaudhary et al., 2012, Al-Bashish et al., 2011). For example, Pixia and Xiangdong implemented several leaf image processing and recognition technologies to study the cucumber disease of downy mildew, powdery mildew, and anthracnose (Pixia and Xiangdong, 2013). Kulkarni et al. used the Gabor filter to extract features and applied artificial neural networks (ANN) to classify leaf images (Kulkarni and Ashwin Patil, 2012). Revathi et al. used a homogeneous pixel counting technique for cotton disease recognition (Revathi and Hemalatha, 2012). Al-Tarawneh applied auto-cropping segmentation and fuzzy C-means classification to analyze leaf color in the olive leaf spot disease (Mokhled, 2013). Kanjalkar and Lokhande proposed to use leaf features extracted from the segmented disease region for disease detection (Kanjalkar and Lokhande, 2014).
Despite the considerable efforts, the existing crop disease recognition methods are not adequate or sufficient due to two limitations: (1) the features extracted from the leaf images of crop disease are in general sensitive to the illumination, orientation and scaling of the images, therefore a preprocessing step prior to feature extraction is often required to account for various translation, rotation, and scaling factors; (2) the features selected for classification are usually treated as equally important, regardless of their actual roles in the classification process. In reality, some features may have little contribution or even have confounding effect to the disease recognition. Assigning equal weights to all the features tends to result in inflated classification error.
SR based classification (SRC) or sparse coding aims to search for the most compact representation of the input sample in terms of linear combination of a small number of elementary samples called atoms, which are usually selected from an over-complete dictionary (Wright et al., 2009). As a state-of-art machine learning technique, in recent years, SRC has been widely used in signal analysis, radar signal formation, image/video compression and reconstruction, and so on. Typical applications include face recognition, human movement recognition, tumor classification, and plant species identification (Wagner et al., 2012, Gkalelis et al., 2008, Zheng et al., 2011, Jin et al., 2015). In this paper, we propose to use SRC for cucumber disease recognition. Noticing that the disease class can be determined by its color and shape features, we combine the log spectrum of the color histogram and the lesion shape of the lesion images as our classification features, and recognize the cucumber disease according to the sparse coefficients. Our proposed method differs from the classical cucumber disease recognition methods in that the recognition is based on both shape features and color features, and the classification features receive unequal weights.
The rest of this paper is organized as follows. Section 2 introduces SRC. Section 3 describes the general procedure of lesion segmentation for cucumber leaves. In Section 4 we propose a recognition method for cucumber diseases based on SR. Experimental results and analyses are given in Section 5. Section 6 concludes the paper and points out our future work on improving the cucumber disease recognition algorithm.
Section snippets
Sparse representation based classification
SR based classification (SRC) method assumes that the training samples from a single class lie on a subspace (Wright et al., 2009, Wagner et al., 2012). As a result, the test sample can be represented as a sparse linear combination of the training samples. This representation is naturally sparse if the number of the training classes is reasonably large.
Suppose that we collect training samples which belong to classes, where , and generally . Let be the
Cucumber leaf image lesion segmentation
Lesion segmentation of cucumber leaf images is a key process in disease recognition (Jaware et al., 2012, Arivazhagan et al., 2013). Typical algorithms such as the K-means clustering can be applied to leaf lesion segmentation, which tries to classify pixels into K classes based on a set of features (Valliammal and Geethalakshmi, 2012, Chitade and Katiyar, 2010). In this paper, we use the K-means clustering algorithm to segment lesion from diseased leaf images. The steps are listed as follows.
Cucumber leaf disease recognition based SR
According to the basic principle of SR, we propose a cucumber leaf disease recognition method as follows.
Step 1 Dataset preparation. Segment the lesion from each diseased leaf image by K-means clustering algorithm.
Step 2 Color feature extraction. Divide the lesion image into three L∗a∗b∗ components, extract histograms of the L∗a∗b∗ components, calculate their 128-point fast Fourier transform (FFT), and extract their log frequency histogram features, denoted as , respectively.
Step 3
Experiments and analysis
To evaluate the performance of the proposed algorithm, we conduct extensive experiments using a leaf image database for cucumber diseases, and compare with four feature extraction based crop leaf disease recognition methods: Support Vector Machines (SVM) (Rumpf et al., 2010), K-means-based segmentation followed by neural-network-based classification (KMSNN) (Al-Bashish et al., 2011), texture feature (TF) based classification (Arivazhagan et al., 2013), and plant leaf image (PLI) based
Conclusions and future works
Fast, automated, image-based crop disease recognition plays an important role in crop disease management. Due to the irregularity, complexity, and diversity of diseased cucumber leaves, many existing classifiers cannot meet the needs of an automated cucumber disease recognition system. Using SR, a newly developed but widely used data representation model, we propose a novel crop disease recognition method based on cucumber leaves. The proposed method uses combined shape and color features from
Acknowledgement
This work is partially supported by China National Natural Science Foundation under grant Nos. 61473237 and 61309008. It is also supported by the Shaanxi Natural Science Foundation Research Project under grant No. 2014JM2-6096. The authors would like to thank all the editors and anonymous reviewers for their constructive advices.
References (27)
- et al.
An image processing based algorithm to automatically identify crop disease visual symptoms
Biosyst. Eng.
(2009) - et al.
Early detection and classification of crop diseases with support vector machines based on hyperspectral reflectance
Comput. Electron. Agric.
(2010) - et al.
Detection and classification of leaf diseases using K-means-based segmentation and neural networks based classification
Inform. Technol. J.
(2011) - et al.
Fast and accurate detection and classification of crop diseases
Int. J. Comput. Appl.
(2011) An Empirical investigation of olive leave spot disease using auto-cropping segmentation and fuzzy C-means classification
World Appl. Sci. J.
(2013)- et al.
Detection of unhealthy region of plant leaves and classification of plant leaf diseases using texture features
Agric. Eng. Int.: CIGR J.
(2013) Nexact interior-point method
J. Optimiz. Theory Appl.
(1998)- et al.
Color transform based approach for disease spot detection on crop leaf
Int. J. Comput. Sci. Telecommun.
(2012) - et al.
Color based image segmentation using k-means clustering
Int. J. Eng., Sci. Technol.
(2010) - et al.
An overview of the research on crop leaves disease detection using image processing techniques
IOSR J. Comput. Eng. (IOSR-JCE)
(2014)
Sparse human movement representation and recognition
IEEE 10th Workshop Multimedia Signal Processing
Detection of diseases on cotton leaves and its possible diagnosis
Int. J. Image Processing (IJIP)
Early detection and localization of sugar beet diseases: new approaches
Gesunde Pfianzen
Cited by (229)
1-Norm twin random vector functional link networks based on Universum data for leaf disease detection
2023, Applied Soft ComputingAutomatic cotton spider mites' damage grading algorithm for depth residual network based on transfer learning
2023, Computers and Electronics in AgricultureDetection method for tea leaf blight in natural scene images based on lightweight and efficient LC3Net model
2024, Journal of Plant Diseases and ProtectionA survey on plant diseases detection using different ML / DL techniques
2024, AIP Conference Proceedings