Bangkit Capstone Project C22-PS152 Documentation C23-PS179

This repository contains the documentation for FarmKu Project.

FarmKu - A platform that focuses on the collection, management, and utilization of farmer activity data as a recommendation for parameters of farming activities from upstream to downstream

Table Of Content

1. Overview
2. How to run?
   • Machine Learning Documentation
   • Mobile Development Documentation
   • Cloud Computing Documentation
3. Team Member of C23-PS179

1. Overview

FarmKu is a project to address challenges faced by farmers in Indonesia regarding soil conditions and plant diseases. The project offers plant disease detection and fertilization recommendations through a mobile application. The plant disease detection feature allows farmers to take photos of leaves that indicate certain diseases. The application uses machine learning models created with TensorFlow to predict the type of disease and provide appropriate treatment options. The models have high accuracy rates, exceeding 90% for most plants. The fertilizer recommendation feature collects soil nutrients data periodically using an IoT device and predicts the soil condition in the near future. If the soil is predicted to be suboptimal, farmers receive immediate notifications to apply fertilizers, helping optimize crop growth. But FarmKu is temporarily pivoting this feature into data collection due to a lack of data.

The project involves three key teams: machine learning, mobile development, and cloud computing.

The machine learning team focuses on developing plant disease detection models using TensorFlow. They leverage transfer learning for leaf detection and disease identification based on plant types. The models achieve high accuracy rates, contributing to accurate disease detection. The models are deployed on a server using Flask, ensuring seamless integration and interaction.

The mobile development team handles the entire application development lifecycle. They start by wireframing and defining the project scope, goals, and user needs. Detailed wireframes and high-fidelity mockups are created, adhering to branding guidelines and UX best practices. The team ensures a user-friendly and visually appealing mobile application.

The cloud computing team plays a significant role in integrating the cloud infrastructure with machine learning capabilities. They establish a robust environment for deploying and scaling the ML models using advanced cloud technologies. Well-defined APIs and service endpoints enable seamless communication and real-time data exchange between the mobile application and the ML backend, enhancing performance and the user experience.

2. How To Run

This API has been deployed using google cloud using cloud run and compute engine you can access it here: ML-Image Process API = http://34.101.59.1:3003/preprocess Main API = https://farmku-backend-hwmomiroxq-et.a.run.app/ Data Collection API = https://farmku-api-hwmomiroxq-uc.a.run.app/

Machine Learning Documentation

If you want to run it locally, u need to do these kind of things:

1. Clone this repository, use ML Folder
2. Open terminal and create flask virtual environment using python -m venv <name of environment>
3. Install all the requiredment with pip install -r requiredments.txt
4. Run full_preprocess.py in terminal to run flask
5. Use url http://127.0.0.1:500/ to access api endpoint.

Machine Learning Documentation: The process of All Data Preparation and Modelling can be accessed in this github repository (you can accessed it after cloning)

1. Load Dataset The Leaf Dataset and our final model already upload in this link (bit.ly/DatasetAndModel). The rest dataset could be downloaded here: Rice Diseases Dataset https://www.kaggle.com/datasets/minhhuy2810/rice-diseases-image-dataset Potato and Corn Diseases Dataset https://data.mendeley.com/datasets/tywbtsjrjv/1 Tomato Diseases Datase https://www.kaggle.com/datasets/kaustubhb999/tomatoleaf Mango Diseases Dataset https://data.mendeley.com/datasets/hxsnvwty3r/1

2. Do Data Preprocessing for All Image Dataset We used ImageDataGenerator library to do data preprocessing, such as rescale, rotation_range, width_shift_range, height_shift_range, shear_range, zoom_range, horizontal_flip, and fill_mode . Training data and testing data have their own preprocessing needs. We only need to rescale data testing into the size that we need.

3. Modelling We use transfer learning to build the model. The Major Dataset is modelled by Inception Architecture, while Leaf Dataset using VGG16 for better result and size.

4. Testing Prediction After the model is ready, we have to test the model to evaluate the result.

5. Save and Convert Model Into H5 Format After the model is ready and evaluated correctly, we could save the model into H5 format. It will enable the model to load by Tensorflow load_model and store it in variable. It will allow us to integrated the model with the API.

Mobile Development Documentation

1. Splash Screen: A captivating screen that welcomes users with an animated logo and a sleek transition into the app.

2. Onboarding: A visually stunning introduction that guides users through the key features of our app using eye-catching graphics and smooth animations.

3. Login Page: A stylish login interface with a modern design and intuitive user experience. Users can securely sign in using their credentials or social media accounts.

4. Register Page: An attractive and user-friendly registration form that allows new users to create an account effortlessly. We've added cool animations to make the process more engaging.

5. Forgot Password Page: A sleek and user-centric interface where users can reset their forgotten passwords easily. We've implemented a secure verification process for account recovery.

6. OTP Page: A cutting-edge one-time password verification page with a minimalist design and smooth transitions. Users receive a unique code via email or SMS for secure authentication.

7. Main Page: The heart of our app, featuring four dynamic fragments:

    a. Home Page: A visually appealing and interactive interface where users can explore personalized content, latest updates, and important announcements.
    b. News Page: A captivating news feed with stunning images and smooth scrolling. Users can stay up-to-date with the latest industry news and trends.
    c. List Land Page: A comprehensive list of land properties with detailed information and high-quality images. Users can easily browse and search for their desired properties.
    d. Profile: A sleek and customizable profile page where users can manage their personal information, profile picture, and app preferences.
   

8. Disease Detection Page: An innovative feature that uses advanced algorithms and image recognition to detect diseases in plants. Users can simply upload a photo of the affected plant for instant analysis and recommendations.

9. Data Soil Land Collection Page: An efficient and user-friendly interface for collecting and analyzing soil data. Users can input relevant parameters and generate detailed reports for land assessment.

10. Record Page: A convenient page for users to keep track of their activities, such as plant care, fertilization, and watering schedules. We've added a visually pleasing calendar view for better organization.

11. Measurement Page: A smart and intuitive tool for measuring various aspects of plants, such as height, width, and growth rate. Users can easily record and track their plant's progress.

Dependencies used during development:

1. Retrofit: A powerful networking library for seamless communication with APIs, enabling smooth data exchange between the app and server.
2. OkHttp3: A reliable and efficient HTTP client that optimizes network requests, ensuring fast and secure data transmission.
3. LiveData: A lifecycle-aware component from the Android Architecture Components, providing real-time data updates and automatic UI synchronization.
4. Glide: An impressive image loading and caching library that optimizes image rendering and improves overall app performance.
5. Navigation Component: A feature-rich navigation framework that simplifies the implementation of app navigation and enhances user flow.

Tools used in the development process:

1. Postman: A versatile API testing and documentation tool that helps developers analyze and debug API requests and responses effectively.

2. Android Studio: The official Integrated Development Environment (IDE) for Android app development, providing a comprehensive set of tools, emulators, and debugging features to streamline the development process.

With these enhancements, our Android Application delivers an even more captivating user experience, combining stunning visuals, smooth animations, and cutting-edge features.

Cloud Computing Documentation

1. Creating Flask App to load model from Machine Learning

• Create simple flask api with the name preprocess.py
• save model and dataset for Machine learning in same directory as preprocess.py
• Load the model in preprocess.py
• create endpoint and test model by running flask using python preprocess.py to run it locally and getting predicted data using local ip.

2. Creating Login and Register with Authentication in NodeJs

• Creating simple Login and Register using MySQL
• create Json Web Token(JWT) to authenticate login and register
• create JWT requirement to request prediction
• change dummy database to cloud sql database
• Test database to user login and register
• Test authentication JWT using POSTMAN

3. Google Cloud Deployment

• create Dockerfile and requirement.txt to store depedency and place it in root directory
• clone api repository in cloud shell
• run this command to deploy

  gcloud run deploy farmku-backend \
  --image gcr.io/$GOOGLE_CLOUD_PROJECT/farmku-backend \
  --platform managed \
  --region asia-southeast2 \
  --allow-unauthenticated
  

• Enable Cloud Run
• select farmku-backend image container and deploy to cloud run

4. ci/cd pipeline with Cloud Build

• Enable Cloud Build
• create cloudbuild.yaml and write command to build new docker container,push it to container registry and run it everytime it trigger
• open CloudBuild and select repository and cloudbuild.yaml as config
• select trigger to everytime push happen in main branch
• build cloudbuild trigger
• add permission to cloudbuild service acccount and run the trigger to automate deployment

5. Create Documentation using Postman

3. Team Member of C23-PS179

ID	Name	University	Learning Path
C340DKX3998	Rayhan Emillul Fata	Universitas Negeri Sebelas Maret	Cloud Computing
C340DKX4157	Samuel Steven P. H.	Universitas Negeri Sebelas Maret	Cloud Computing
M038DKX4066	Benedictus Kenny T.	Institut Teknologi Sepuluh Nopember	Machine Learning
M351DKX4183	Alvin Fernando S.	Universitas Surabaya	Machine Learning
A351DKX4108	Victor Manuel S.	Universitas Surabaya	Mobile Development
A351DKX4106	Michael Andreas	Universitas Surabaya	Mobile Development