AgriDet: Plant Leaf Disease severity classification using agriculture detection framework

https://doi.org/10.1016/j.engappai.2022.105754Get rights and content

Abstract

In the field of modern agriculture, plant disease detection plays a vital role in improving crop productivity. To increase the yield on a large scale, it is necessary to predict the onset of the disease and give advice to farmers. Previous methods for detecting plant diseases rely on manual feature extraction, which is more expensive. Therefore, image-based techniques are gaining interest in the research area of plant disease detection. However, existing methods have several problems due to the improper nature of the captured image, including improper background conditions that lead to occlusion, illumination, orientation, and size. Also, cost complexity, misclassifications, and overfitting problems occur in several real-time applications. To solve these issues, we proposed an Agriculture Detection (AgriDet) framework that incorporates conventional Inception-Visual Geometry Group Network (INC-VGGN) and Kohonen-based deep learning networks to detect plant diseases and classify the severity level of diseased plants. In this framework, image pre-processing is done to remove all the constraints in the captured image. Then, the occlusion problem is tackled by the proposed multi-variate grabcut algorithm for effective segmentation. Furthermore, the framework performs accurate disease detection and classification by utilizing an improved base network, namely a pre-trained conventionally based INC-VGGN model. Here, the pre-trained INC-VGGN model is a deep convolutional neural network for prediction of plant diseases that was previously trained for the distinctive dataset. The pre-trained weights and the features learned in this base network are transferred into the newly developed neural network to perform the specific task of plant disease detection for our dataset. In order to overcome the overfitting problem, a dropout layer is introduced, and the deep learning of features is performed using the Kohonen learning layer. After percentage computation, the improved base network classifies the severity classes in the training sets. Finally, the performance of the framework is computed for different performance metrics and achieves better accuracy than previous models. Also, the performance of the statistical analysis is validated to prove the results in terms of accuracy, specificity, and sensitivity.

Introduction

Agriculture is the most primary and essential source of furnishing national income based on a country’s quality and quantity of products, especially crops and plants. According to the report of the Ministry of Agriculture and Farmers Welfare (MAFW) community, 58% of people primarily depend on agriculture for their occupation, mainly in India. Natural disorder factors such as pests, weeds, and diseases account for 15%–25% of crop production losses. Monitoring of disease plays a significant part in the successful development of crops on the farm with the support of efficient farming procedures. In the beginning, a specialist in that sector would manually monitor and analyse plant illnesses. Several monitoring-based devices are used in plant disease detection, but this adds complexity and raises hardware costs. Hence, farmers with limited income cannot afford to buy such monitoring devices for detecting diseases. Vishnoi et al., 2020, Sharath et al., 2019, Ale et al., 2019. To provide a reliable treatment for the disease, the diagnosis of leaf diseases must be incorporated into the tools. Therefore, raising crops for the livelihood of the people and increasing their productivity are the essential goals of a farmer. The growth and yield of the crops are adversely affected due to plant diseases. Hence, the diagnosis of plant diseases is a major concern for the management and production of crops (Barbedo, 2019, Thomas et al., 2018, Shruthi et al., 2019). This involves a huge amount of labour and takes a long time to process. Techniques for image processing can be used to determine plant diseases. Disease symptoms are often visible on the fruit, stem, and leaves. The plant leaf is taken into consideration for disease identification as it exhibits disease symptoms.

Numerous methods have been introduced by the researchers to achieve better results in identifying the types of diseases in the leaves. In earlier days, manual detection methods were used in the prediction of crop diseases. This is where experts classify the disease and recommend a treatment plan. This process becomes challenging due to the high cost of labour as the disease detection process is difficult and has a high computational overhead. Golhani et al., 2018, Sandhu and Kaur, 2019. The image-based detection and classification methods were determined based on Plant Leaf Disease (PLD) measures. This reduces the labour costs as well as the processing complexity (Radovanović and Đukanovic, 2020, Indumathi et al., 2017, Nafi and Hsu, 2020, Prathusha et al., 2019). Therefore, an automatic image-based method for disease classification needs to be developed for agricultural applications. Many state-of-the-art algorithms in image processing, such as deep learning, neural networks, Support Vector Machine (SVM), and Scale Invariant Feature Transform (SIFT), are used for plant leaf disease detection and classification. But most of these systems are inaccurate and involve millions of parameters for training the images to provide classification (Tete and Kamlu, 2017, Goncharov et al., 2018, Sardogan et al., 2018). Also, convolutional auto-encoder–decoder-based approaches are used for detecting plant diseases. Further, the system incorporates hybrid techniques to detect the disease spot in the leaves. These models, however, suffer from multiple complex backgrounds, complex processes in training huge datasets, a lack of multi-scale features, misclassified patches caused by samples being affected by environmental factors, and a high execution time due to the large training set. Hence, there is a need to develop an efficient, cost-effective image processing system for accurate detection and severity classification of plant diseases.

Motivation

In the current era, plant disease detection has become a major focus in agricultural applications involving the improvement or modification of soil, crops, livestock, poultry, fish, or shellfish and their resulting products as they relate to human health using machine learning models (Ramesh et al., 2018, Daniya and Vigneshwari, 2019). Further, it relates to bacteria, fungi, and viruses, as well as pest organisms. This detection model will make it easier for many farmers to detect plant diseases early, prevent plant waste, and protect disease transmission from diseased to healthy plants. Several existing studies on plant disease detection have used network models such as GoogleNet, AlexNet (Jadhav et al., 2021), ImageNet, or VGG19 (Chen et al., 2020). However, these models cannot update the network, and the higher computational cost and network structure were relatively high. Furthermore, automated systems based on the deep learning model (Chohan et al., 2020) have several issues, including complex multiple backgrounds, complex processes in training massive datasets, a lack of multi-scale features, and extensive knowledge. As a result, failure to detect and classify plant disease stage level is a major problem in agriculture. Therefore, the existing methods have gaps in satisfying the above requirements for image detection tasks from plant disease images. Motivated by these facts and existing studies, the AgriDet framework is proposed in this paper to accurately detect plant disease and classify the stage level of plant disease with high accuracy.

The major contributions of the proposed AgriDet framework is presented as follows,

  • The AgriDet framework is designed for detecting diseases in plants through image processing techniques. This incorporates the conventional INC-VGGN as the base network with a new neural network architecture built for disease prediction.

  • The proposed framework performs pre-processing of images to eliminate the issues in image acquisition, and further, a multi-variate grab-cut algorithm is proposed for image segmentation to tackle the occlusion problem when considering multiple backgrounds in images.

  • The Kohonen-based deep learning is performed to more accurately learn the multi-scale features of the leaf disease by using the previously learned features in the INC-VGGN model to improve the detection accuracy. Further, overfitting is avoided by using the dropout layer in the network.

  • Lastly, a pre-trained model based on conventional INC-VGGN is employed to construct a Quantifying Disease Severity Classification (QDSC) strategy that extracts the pre-learned features and classifies the disease severity classes.

  • The performance of the proposed method is compared with other existing approaches under different performance metrics.

The structure of the paper is organized as follows: In Section 2 the literature survey is presented. Followed by this the working process of the AgriDet framework is explained in detail in Section 3. Section 4 presents the performance evaluation of the proposed work. The applicability of the proposed approach in real-time testing and detection is discussed in Section 5. Finally, Section 6 provides the overall conclusion, limitations and future works.

Section snippets

Literature survey

In this section, a survey of various existing plant disease detection techniques is presented to better understand their workings, and flaws are discussed as follows:

With the rapid growth of the population, agriculture is an essential part of all countries’ energy sources. However, plant diseases have an influence on the quantity and quality of crops for agricultural expansion. In order to prevent and control plant diseases, it is crucial to first diagnose them. The approach used for plant

Proposed AgriDet framework

In the proposed framework, we have proposed an Agriculture Detection (AgriDet) framework that incorporates conventional INC-VGGN and Kohonen-based deep learning networks to detect plant diseases and classify the severity level of diseased plants. Initially, the preprocessing step is performed to eliminate the unequal and improper size of the images using image scaling, enhancement, and contrast adjustments. To solve complex multiple background problems, the Multi-variant Grabcut algorithm (MGA)

Experimental results and discussion

This section comprises of details about the experimental settings, dataset used, metrics, visualization results and comparison results.

Discussion

The vital purpose of the AgriDet framework is to support the farmers in getting a clear result on the type of plant disease that has occurred in the field. During real-time testing, any farmer whose field is affected by the plant disease captures images of the crop. This image is sent as input into the AgriDet framework, where disease detection is performed. Initially, the image is subjected to pre-processing and segmentation. Then, the learned features from the pre-trained model are employed

Conclusion

This paper presents an AgriDet framework that incorporates conventional INC-VGGN and Kohonen-based deep learning networks to detect plant diseases and classify the severity level of diseased plants. Initially, the preprocessing step is performed to eliminate the unequal and improper size of the images using image scaling, enhancement, and contrast adjustment samples collected from the PlantDoc and PlantVillage datasets. To solve complex multiple background problems, a segmentation process is

Funding

There is no funding for this study.

Ethical approval

This article does not contain any studies with human participants and/or animals performed by any of the authors.

CRediT authorship contribution statement

Arunangshu Pal: Conception and design of study, Material preparation, Conceptualization, Data collection and analysis, Methodology, Formal analysis, Investigation, Writing – original draft, Writing – review & editing. Vinay Kumar: Conception and design of study, Formal analysis, Investigation, Supervision.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

All authors approved the version of the manuscript to be published.

References (43)

  • DaniyaT. et al.

    A Review on Machine Learning Techniques for Rice Plant Disease Detection in Agricultural Research System. Vol. 28. No. 13

    (2019)
  • GoncharovP. et al.

    Disease detection on the plant leaves by deep learning

  • He, K., Zhang, X., Ren, S., Sun, J., 2016. Deep residual learning for image recognition. In: Proceedings of the IEEE...
  • Huang, G., Liu, Z., Van Der Maaten, L., Weinberger, K.Q., 2017. Densely connected convolutional networks. In:...
  • HughesD. et al.

    An open access repository of images on plant health to enable the development of mobile disease diagnostics

    (2015)
  • IndumathiK. et al.

    Intelligent plant disease detection system using wireless multimedia sensor networks

  • IniyanS. et al.

    Plant disease identification and detection using support vector machines and artificial neural networks

  • JadhavS.B. et al.

    Identification of plant diseases using convolutional neural networks

    Int. J. Inf. Technol.

    (2021)
  • KarthikR. et al.

    Panoramic view creation using invariant momentsand SURF features

  • KarthikR. et al.

    Attention embedded residual CNN for disease detection in tomato leaves

    Appl. Soft Comput.

    (2020)
  • KhanM.A. et al.

    An automated system for cucumber leaf diseased spot detection and classification using improved saliency method and deep features selection

    Multimedia Tools Appl.

    (2020)
  • Cited by (38)

    • Composite descriptor based on contour and appearance for plant species identification

      2024, Engineering Applications of Artificial Intelligence
    View all citing articles on Scopus
    View full text