In this article, I will share the tools that I have used to build the plant disease identification app (available on Google playstore) and steps right from the start to its deployement.

I am using flutter because of its fast development cycles and ability to built fast and smooth UI. I recommned reading flutter documentation.

Make a new flutter project on Android studio (or whatever IDE you use) and add these plugins in your pubspec.yaml file.

1. Camera plugin: This plugin will enable the app to capture images and stream image buffers to dart. We need this plugin to capture plant leaf image and then use our tensorflow model to give results. Plugin: link

2. TensorFlow lite plugin: Tensorflow is open-source library for machine learning. I will use tensforflow specifically for model training. You can learn more about tensorflow on their website. You can use teachable machine to first create a demo machine learning model and then convert it to tensorflow lite so that you can check whether your app is displaying results as intended. Plugin link

3. Cached Network Images: The plugin cached network image will load and cache network images. Although it will make the application internet dependent but it will significantly reduce the size which is very important to compete with other apps on the Google playstore. Your download size matters a lot while convincing users to hit that install button. I hosted my images on imgbb.com, the loading times are pretty fast. Best thing about this plugin is that it caches the images once they are loaded. Next times user visits the same page, it feels like image is loaded instantly. I made the decision to use this plugin because I really want to reduce the size the app. Plugin link

4. Firebase core: It connects the app with multiple firebase services. I am using this to display articles from the firebase database. In future updates, I will enable authentication so that I can store profile specific user data to add more functionalities in the app. To use this plugin first you have to create a new application in firebase. If you are enabling authentication in the app then you have to enter you SHA key while setting up the app or you can do it after creating the app in the settings section. To get SHA fingerprint from android studio go to this stackoverflow link. I strongly recommend reading about the basics of firebase and the rules to setup database on their website.

Front-end thought process:

I used figma to just get an idea of how my app will look like. I used horizontal Tab bar to switch between categories because it is similar to whatsapp. Majority of the interenet using population have installed whatsapp on their phone so they are familiar with the UI of whatsapp. Even the old people won’t be intimidated while navigating through the app. Here is the link for implementing the tab bar. The smooth transition from homepage to a specific plant page is done with Hero animations. It looks so cool, I use it for almost every time I have to make change pages with a common asset between them. To get the photo of the fruits, vegetables and plants, I used pixabay.com and pexels.com because they have plenty of high quality images to choose from for free.

I compressed these images. why? Even though I am using cached network images to load all these images from web and it is not effecting the app size. But I want to decrease the load time of the photos that’s why I compressed these images using compressnow.com

I cropped their background using this online AI tool photoscissor.com. It will look good because we are using Hero animation.

I used same online tools that I have mentioned above to create background image of leaf. I inverted this image horizontally to use it on second tab bar background to make switching between two tab bars feel more natural.

Back-end and Tensorflow lite:

For backend, I use NoSQL cloudfirestore database to display articles on the app. There is no authentication in my app so I didn’t tweaked any default rules of database except some basic changes to make the app production ready. If you want to add authentication to your app, you can consider using authentication feature of the firebase. You can authenticate your users using Google account, Facebook account, Apple account, OTP on their phone, Email etc. You can even send email verification link to your user. It is super simple, reliable and fast. I used TensorFlow lite because I want the machine learning functionality of my app to work offline. Using Tensorflow lite instead of deploying a model on server(you can also do this using firebase) is a tradeoff between accuracy. Although Tensorflow lite model is less accurate but it is enough for classifying between diseased and undiseased plant leaf. I got data from Kaggle, open source data website, and agriculture universities that have repository of data. My university professor helped me a lot in finding the data and connecting me with a team who did the similar model training that I am going to do. It was very learning and fun experience.

I programmed some simple java helper classes to establish communication between tensoflow lite model and my application. The tflite plugin of flutter has some limitation that I hope they will address in future updates.

Deploying on Google Playstore:

I already had an account on Google playstore to publish the app. It is just one time fee of I think 20 something dollars. For privacy, I used free generic terms and conditions tool. I hosted my terms and condition on my website(fouxa.com) which is hosted on netlify. If you think your app needs custom terms and condition then you should contact a expert. There are lots of online services that can help you with terms and conditions of your app.

I created screenshots and App Icon using Figma. You have to follow these guidelines while creating screenshots, App Icon or videos. I also edited a video to give a brief description of the app. I created this video using free clips available online(same sites that I have mentioned before to get free images) and edited them using da vinci video editor.

If you have any questions or having any difficulty building similar stuff, contact me through social media.