Automated Disease Detection And Analysis Of Grape Leaves
issue 1

Automated Disease Detection And Analysis Of Grape Leaves

Shafique Mohammad¹, Robin Tiple², Hemant Dindewar³, Pratik Bobhate⁴,Prof. Manisha Pise5

1,2,3,4B.E students, Department of Computer Science and Engineering, Rajiv Gandhi College Of Engineering Research And Technology, Chandrapur, Maharashtra – 4424015Professor, Department of Computer Science and Engineering, Rajiv Gandhi College Of Engineering Research And Technology, Chandrapur, Maharashtra – 442401

Abstract— Agriculture is one of the traditional and most important occupations of man which is required for sustaining life. Hence, many countries pour large amounts of money into research on new techniques for efficient and cost-effective farming. This is the reason it is important to detect and classify diseases of the plants in the early state to ensure that the effect of disease on the crop can be mitigated.

If we are not able to detect the disease in the early state then it would mean huge losses in product quality, quantity, or productivity is affected.

However, doing this observation of detecting diseases by farmers manually means a lot of man-hours wasted only on observation and doing nothing productive. Also, such observation also required expert vision to understand the disease which is only possible if you are an expert in this field.

That is why we need fast automated learning and classifying technique or a system that can do this work for us in a very cost-efficient manner. This paper presents a solution for that which uses CNN(Convolutional neural network) and a simple web application that uses it to get the leaf image and classify whether it is healthy if not, the exact disease it has been affected with.

Keywords: Deep learning, neural networks, convolutional neural networks, CNN, image segmentation

1. INTRODUCTION

Agriculture is one of the key development areas in human civilization which enabled people to live in cities with a surplus stock of food. Large economies like China, India, the United States of America and Brazil constitute about 59.56% of the total agricultural production in the world, as of 2016. Hence it is necessary to make the most efficient farming and get the highest quality and quantity product. However, leaf diseases are one of the main concerns for the farmers which not only degrades the quality of the product but also if ignored for some time will also damage it a lot that you have to throw most of it away. However, at the same time it is not possible for a farmer to continuously monitor, observe the farm and leaves for possible diseases. Such manual observation requires a  lot of man-hours, also a normal farmer is not an expert to detect it just with a glance.

That is why research on disease detection of crops is getting attention nowadays. A Grape is a berry fruit of the plant family Vitis. It is a majorly consumed fresh fruit in India. It is packed with nutrients, especially vitamins C and K. Also they can be used for grape seed oil, raisins, jelly, wine, jam, grape juice, vinegar, etc. The area under grape farming in India is about 79.6 thousand hectares with annual production of 1878.3 thousand metric tonnes.

Being an important cash crop, the livelihood of millions of people depends on the production of grapes each year. So, it is important to prevent any type of disease from forming on the grape plant. It is observed that for this purpose farmers still do naked eye observation. Though with years of observation farmers have an eye for catching the diseases it is still not feasible for them to monitor such large areas with only limited workers and time. Hence, they might want to call experts to help them out. But that tends to be very expensive which in turn will increase the cost of the produce, and it is not desired. They need an automated and cost-effective system that will help them detect the diseases in the early stage so that they can apply curative as well as preventive measures. Some of the diseases of the grape plant are namely, Fungi Bacteria Viruses Nematodes, Anthracnose, Bitter Rot, Black Rot, Botrytis, Downy Mildew, Eutypa Dieback, Dead Arm, Macrophoma Rot.

Deep learning systems combined with computer vision and image segmentation techniques will greatly help solve such problems and be very beneficial to the farmers. Computer vision systems are already being used in the agriculture and manufacturing industry for sorting of fruits in fruit processing, classification of grains, medicinal plant identification,etc. In all the above examples, the implementation of computer vision is done by taking digital images, taken using digital cameras and then image processing is used to extract required and important features that are necessary for additional analysis.

The same process and techniques are used in the disease detection of the grape plants.

2. LITERATURE SURVEY

Using Image Processing to convert the grape leaf images into something that can be processed effectively by the system [1]. Grape leaf images that are taken for inspection by the farmer can not be directly processed by the classifier and needs proper pre-processing. This is done to increase accuracy. The accuracy is increased by discarding some parameters of the leaf images and enhancing the other useful parameters.

The main four steps in the pre-processing scheme are, first, a color transformation of the RGB formatted image, then masking the green color pixels that have a high probability of consisting of the leaf in the image. Thirdly, use these masks of the pixels to classify and create segments on the image. And finally, use segmentation techniques to properly identify the leaf region so that the classification is only done on the actual leaf part and the rest is discarded.

Identify and extract features of the disease infected region on the leaf and compare them to the healthy and other classes of diseases [2].  

It is observed that the accuracy by which diseases are classified is significantly increased after using the segmentation of the leaf image and properly identifying the leaf region.

Furthermore, you could also segment the leaf disease infected region of the training images to catch even the minute similarities that may be useful in training the model. After using the segmented data it has been found that even a simple CNN classifier recorded about 97% accuracy jump from 93% on the validation set after proper image processing and extracting proper texture features. Moreover, this number can be increased further by using other complex models that are already available.

Study of the state of the art Image Processing technique[3]. Various state of the art image processing techniques are already available that can be used directly. Some of these techniques that could b used include The state of the art review of different methods for leaf disease detection using image processing techniques is presented in paper[3]. The present methods studies are for boosting throughput and reduction subjectiveness which happens due to naked eye observation through which identification and detection of plant diseases are done.

A survey of plant leaf disease detection techniques[4]. Different symptoms of the four diseases of the grape leaves are studied and identified in this paper. Some of the available techniques to identify the diseases are also highlighted in the paper. The proposed system will increase the effectiveness of disease identification and also alert the farmer as early as possible.

Simple and user-friendly leaf disease detection as a service. The system proposed in this paper gives an idea of how digital space can be effectively used to create and manage the leaf disease detection by a simple and user-friendly management service.

Plant disease recognition based on deep learning techniques. The model of the disease classifier proposed in this paper uses CNN(Convolutional Neural Network) model and other deep learning techniques such as data augmentation, image segmentation while training and classifying that use cutting edge tools to improve the efficiency of the solution.

3. PROPOSED SYSTEM ARCHITECTURE

The architecture of the system proposed is depicted in the image below. The main objective of this intelligent system is to facilitate the farmers into managing the crop diseases without much manual labor into its observation. This reduces the overall losses. This system is designed in such a way that efficiency is maximum.

The basic idea is to first obtain a dataset with already classified images of grape leaves which can then be used by a CNN model to train itself. And then, use this model to classify the images uploaded by the user. Classification is done by comparing the image uploaded by the user and the images that are already present. This comparison is not done directly on the images but on the mathematical model that is created by the deep learning models and techniques, basically the features that are extracted.

All the stages of the system are explained briefly below.

Also, we have divided the proposed solution into two distinct parts. The first one is to train the model so as to accurately classify the images. The second one is the actual real-time API and web/mobile app which facilitates the users to upload the images and get the results online.

Fig 3.1: Overall structure of the system

We introduce an image-processing and deep learning-based solution for automatic leaf disease detection, classification, and analysis. We have tested our solution on three diseases which affect the grape plants. Hence the four classes of the classifier are

  1. Black rot
  2. Black Measles
  3. Leaf blight(Isariopsis_Leaf_Spot)
  4. Healthy leaf.

The general process is:

Firstly, we obtain or acquire the images of the leaves either by directly uploading the image by the user or get real-time image data that is processed with some time intervals in between by using digital cameras. Then image pre-processing and segmentation techniques are applied to this image to get the proper data from which features can be extracted and analysis can be done.

After that, the classifier can run and use the previously trained model to rigorously classify the real-time data.

3.1 Description of the overall system

  1. User: Here the user is the farmer. He may choose to directly upload images or use a real-time, always-online digital camera that provides the images for observation to the classifier.
  2. Computer System: This is the server system where the model and classifier resides. This server responds to the requests made by the user to properly classify the images given to it.

3.2 System Algorithm

Basic steps for describing the proposed system are as follows:

  1. Image Acquisition.
  2. Image Pre-processing
  3. Segmentation
  4. Feature Extraction
  5. Classify the extracted features.
  6. Get result.

1. Image Acquisition:

During the training phase of the model, the images are acquired using the dataset that is already available online for free.

As discussed above, the images are obtained either by the farmer or a digital camera. If the images are obtained by the digital camera the data needs to be real-time but the images should only be classified with a proper time interval between two consecutive images to not overload the system.

2. Image pre-processing:

This is the second step, where the image is pre-processed for enhancing the image so as to discard all the useless parameters of the image and enhance the required parameters. This stage involves color conversion, masking of the color pixels, histogram equalization, clarity enhancement, etc.

Some complex popular methods for image processing also include Anisotropic diffusion, Hidden Markov models, Independent component analysis, etc.

3. Segmentation:

Also called Image Segmentation. Basically, in image segmentation, a digital image is partitioned into multiple segments which can be defined as super-pixels. The benefit of doing so is that you can identify the leaf region in the image and discard the background noise.

This way, you have a more meaningful representation of the data.

4. Feature extraction:

The features here are the features of leaf images that help classify the leaves into one of the four categories. Some of the features that are taken into consideration are color, texture, shape, and edges, etc.

Extracting the important and useful data from the input image of the grape leaf is also the process of feature extraction. Also converting the input data into the set of features is called feature extraction. So in this proposed system color and texture features are selected to get good results and accuracy.

5. Classification:

Classification is the last stage of the system. It is used for both the training of the model and then classifying the input images.

Classification basically involves the comparison of the input data features to the features that have been learned by the model from the training dataset. Here we use the CNN(Convolutional Neural Network) as the classifier model. This model has many 2D layers. Even with this simple model, it has been observed that we get about 97% accuracy. This number can furthermore be increased by using complex models that are readily available in a large number of deep learning frameworks.

4. RESULT

Here we present some screenshots of the prototype of the system developed that solves the problem of grape leaf disease detection.

Fig 4.1: Web GUI
Fig 4.2: Upload Black Rot Image
Fig 4.3: After Uploading Prediction
Fig 4.4: Predict Disease and give remedies
Fig 4.5: Prediction(Healthy Leaf)

5. CONCLUSION

We have developed an automated leaf disease detection management service with the help of the CNN model which is capable of classifying three grape plant diseases. Proper image processing and image segmentation techniques were applied. The feature set was created to accurately compare and classify the disease. The parameters or features of a healthy and disease-ridden plant were identified. The accuracy rate is about 97% after using image segmentation and data augmentation. Further steps to improve the accuracy were identified which includes using another complex deep learning model and we can also grow the training samples and extract features from leaf texture.

6. ACKNOWLEDGMENT

We would like to express our special thanks of gratitude to our guide Prof. Manisha Pise for her encouragement, able guidance, and valuable inputs. We also express our sincerest regards to our Head of Department Dr. Nitin J. Janwe for encouraging and allowing us to present this work.

7. REFERENCES

  1. Sujatha R*, Y Sravan Kumar and Garine Uma Akhil,” Leaf disease detection using image processing ”, Journal of Chemical and Pharmaceutical Sciences, January – March 2017.
  2. Vijai Singh, A.K. Misra, “Detection of plant leaf diseases using image segmentation and soft computing techniques”, Information Processing In Agriculture 4 (2017) 41–49
  3. Naikwadi Smita, Amoda Niket. Advances in image processing for the detection of plant diseases. Int J Appl Innov Eng Manage 2013;2(11).
  4. Rathod Arti N, Tanawal Bhavesh, Shah Vatsal. “Image processing techniques for detection of leaf disease”. Int J Adv Res Comput Sci Softw Eng 2013;3(11).

Related posts

Defence Mechnisms By Using Decoy Technology

admin

Face Recognition Using Eigen Face

admin

LIE DETECTION SYSTEM USING ARTIFICIAL NEURAL NETWORK

admin

Leave a Comment