Intelligent Plant Disease Detection with Raspberry Pi

The agricultural sector is the backbone of every country because 90% of the population depends on crop production. We discuss image processing techniques to detect leaf diseases and the K-means clustering algorithm for image analysis. This aims to determine the leaf area affected by the disease early and prevent the emergence of larger farms. In agricultural research, leaf disease detection is a necessary subject because it can monitor field cultivation forms. Therefore, through this method, the symptoms of the disease can be automatically identified. This work provides the best way to determine the percentage of leaves affected by the disease, and providers offer early information to farmers through the IoT and Machine Learning system. In this work, we performed this process on rice leaves and tomato leaves simultaneously and determined the percentage of the affected area. The cost of pesticides and other products will be reduced when we improve agricultural productivity. Farmers have many options for planting in the field. However, planting these crops can increase yields and reduce the shortage. Quality is achieved through technical means, so that technology can be used to increase productivity and improve quality. Generally speaking, when plant diseases occur, it can be said that leaves are the leading indicators of plant diseases. These diseases are due to a lack of nutrients and minerals. Some of them are related to climate change, while others are related to bacteria, viruses, and fungi.

In most cases, we can see spots on the leaves due to diseases; however, there are many diseases in plants. By 2050, the world's population will reach 9.6 billion. Therefore, to feed such a large population, agriculture must adopt IoT and Machine learning technology. These conditions need to be met to solve problems such as severe weather conditions and detailed farming methods. This work predicts the leaf infestation and percentage of the spread. The traditional way of disease detection is through identifying and detecting plants and the objective optical observation by experts. This requires a large team of experts, who are still under constant supervision of experts, so when the farm is large, the price is very high. At the same time, farmers in some countries do not have the suitable space or may not have the concept of seeking help from experts because although the price of expert consultants is high, the time required is enormous. In this case, the recommended technique can be used to monitor large areas of land. It is easier and cheaper to automatically detect diseases by simply looking at the symptoms on plant leaves.

Method 1:

We use Raspberry Pi for capturing image and computation process.

The captured image is first sent to through the Raspberry Pi. The trained classifier modeling and classifying the captured image and finally returns the classification output. The result pushes into cloud database which can be retrieved by phone.

The system architecture is the conceptual model that defines the system. The Raspberry Pi will be the system's core that controls the camera and the trained Pi image classifier. Take a photo and save it in a specific file in the Raspberry Pi's internal memory. The newly trained image classifier then starts to classify the image. Finally, users can access data from Firebase through the Android mobile app. The Android mobile app only displays the latest sorted data and sends a notification when the results are not sorted correctly. This module is responsible for training the image classifier for this.

There are two hardware implementations for the Raspberry Pi. The first implementation is the status indicator configuration, which uses 3 of the 40 GPIO pins available on the Raspberry Pi. This is the realization of camera settings using the CSI camera port provided by Raspberry Pi. You also need two different colored LEDs (2 pins). These LEDs indicate the status of the system. Each color of LED shows a different level for a complete system; no user can see the system status screen.

To train the classifier, there must be a module that contains all the necessary files, such as training scripts, a pre-trained model to reduce training time, and an image dataset used for training. All settings of the camera, timer, and LED functions are written and implemented in Python code. This is responsible for configuring the camera parameters of the Raspberry Pi, controlling the timer for taking pictures, saving the images in a specific file, and displaying the status. Automatic restart of the system and the system where the error occurred. Then call the image classifier model to classify the image.

After the classification is completed, display the classification results and sending the classification results to the cloud database. If an error occurs in one of the processes, the system will automatically restart. Once the project is completed, switching from the Raspberry Pi to the cloud database can be automated. The flowchart simulates the situation that the plant disease detection system records every 3 hours, and then the trained image classifier starts to classify the captured images, returns the classification results, and sends the results to the cloud database. Cloud database, users can use the Android mobile phone application developed for this project to access data.

The application compares the date of the ranking result with the current date and displays the latest development. Finally, the results are compared to determine whether it is healthy or unhealthy. If the result is incorrect, a notification will be sent to the user. The purpose of the database is to store the ranking results used in the mobile application. The developed mobile application retrieves the classification data from the database and informs the user that there is a positive result regarding the existence of the disease. It shows the methods to deal with various plant diseases. This function allows users to take correct precautions to eliminate plant diseases and prevent their spread on the farm.

The overall system architecture flow-chart is provided here:

This is how the application will be looking like: