Conversational AI- Teaching Computers to Assist Farmers. Why there is a need for conversational (AI) Agriculture Intelligence system using Voice Assistants?
Farmers and Policy makers in India are desperately looking for smart Agriculture Decision support system to make quick and effective decisions based on #AgData.
Voice Assistant alongside NASA WorldWind Framework, can help amplify and accelerate AgData usage through Machine Learning and have the potential to bring monumental changes in Indian Agriculture space by engaging with farmers in an interactive & conversational way.
We strongly believe digital agriculture, where technologies such as Artificial Intelligence (AI), Machine Learning, Satellite Imagery and advanced analytics can empower and benefit small-holder farmers and decision makers in agrarian economy like INDIA.
The adoption of technology “interactive Voice assistant within Farming ecosystem” could spur a whole new revolution in India and usher in an era of unprecedented productivity and prosperity.
1. Historic Rainfall data from 2000-2017 can be visualised from 10 rainguage stations representing the whole district can be visualized
2. For Quick and Easy interpretation, on top of NASA WorldWind, the user have the option to see the Monthly and Yearly Rainfall in an Interactive way, the User has the Option to see all stations cumulative Rainfall representing the whole district and have a quick review on the district on the Historical Rainfall
3. The User can view the predicted rainfall using Machine Learning, based on Historic rainfall data from 2000-2017 and the codes are open sourced, the user can feed in their own time series data, so that people can predict, visualise and interpret the results.
4. The Live Weather data (Temperature and Humidity) integration is done, if the user click on the stations in the NASA WorldWind, the user can quickly view the Live data
5. The user can look in to relevant satellite imagery like Vegetation Index, Leaf Area Index, Rainfall, and Water vapour & Land surface Temperature to visualize and interpret on top of NASA WorldWind, to know what has happened before in our area of interest.
6. For Conversation AI, the user need to have the Google AIY Developer Kit https://aiyprojects.withgoogle.com/
7. The voice assistance is able to answer questions related to rainfall, Temperature and weather condition at Theni district for past, present and near future in communicative way
8. Unanswered Questions, for which the Voice kit is not trained for, gets automatically stored to Firebase, which can help us build a much more robust conversation AI system for Farmers
Goal of this project is to put conversation AI in to the hands of the farmers and Policy makers, in the proposed prototype, we have showcased how a conversation system between Farmers and the voice Assistant within the ecosystem of NASA WorldWind can help analyze, visualize, interpret and communicate the critical data points related to Agriculture.
We have selected Theni district in Tamilnadu, India.
The app has 3 data sources Historical rainfall data, satellite imagery from NASA & live weather data from openweathermap
Here we show the system architecture of Voice Assistant Interface along with NASA WorldWind Framework in an android app, to start with we are building a Learning system with ground data that can effectively learn, predict and communicate the results to the respective stakeholders.However we are still working on to add Satellite Imagery in the next phase and create an effective decision support system through voice based assistant
Here we run the quick demonstration on how to run the app
In the same app you can view and Interpret, ground data, Live data and Satellite data on top of NASA WorldWind If you click on the Ground Data icon, & select from the pull down menu, Demo(Rainfall), the placemarks of the Ground Rainguage stations will be shown on the NASA WorldWind. Now you can click on the individual stations to know the historical rainfall data from 2000-20017, to help user in easy and quick analysis, we have created a animation of the Rainfall map representing the whole district with the 10 stations.The user has to click on the analysis icon with the play button to play the gif animation of the Rainfall map
If you click on the placemark of the individual rainguage stations, the user can view the Monthly cumulative rainfall from the period of 2000-2017. Using the Radial Basis Function Model on top of the historical timeseries daily rainfall data, we have predicted the 2018 rainfall data.
From the Pulldown menu of the Grounddata the user can select the Temperature and Humidity live data from the open weathermap which can help amlify the region specific AgData Analysis
If you click on the custom data, the app can acess the Smatphone database and the the user can feed in their own time series data, so that people can predict, visualise and interpret the results based on their region of Interest.
The user can look in to relevant satellite imagery like Vegetation Index, Leaf Area Index, Rainfall, and Water vapour & Land surface Temperature to visualize and interpret on top of NASA WorldWind, to know what has happened before in our area of interest.
For Conversation AI, the user need to have the Google AIY Developer Kit https://aiyprojects.withgoogle.com/ withoutwhich the the conversional functionalities will not work.
The voice assistance is able to answer questions related to rainfall, Temperature and weather condition at Theni district for past, present and near future in communicative way. In the Above video we have showcased using the Voice assistant the user can ask the app, the trained questions, like what is the rainfall at a specific station, we have made the Voice assistant respond in 2 different ways
Ressponse 1 is through Voice, if the questions are asked to the voice assistant then it responds in voice in a communicative way (In the Pitch Video you can see the demonstartions of the conversation AI) Response 2 is instructing the app to open those Stations for visualing the results.Basically instructing the app through voice to do some specific tasks ( In the above video u can see the demo of this application)
Unanswered Questions, for which the Voice kit is not trained for, gets automatically stored to Firebase database, so that we can finetune to have the better conversation system .
Our Team Really enjoyed the Field Demonstration
Here are some Takeaways
The Main challenge is, conversational Voice Assistant supports only English, so it’s tough from our side to explain the Farmers to ask questions in English. Then it is a hard time for the Voice Assistant to understand the Farmers Slang of English!!! The Voice Assistant too performed well in understanding most of the Farmers and give the required results in a conversational way.
But
There was so much of Curiosity from the Farmers side to Know about the Voice Assistant, How it works, and engage or communicate with the Voice Assistant
We’re still in the early days of conversational AI, but this field will evolve quickly
We are closely working with PTR College of Engineering and Technology to Leverage each other strengths and accelerate the Farmer Outreach, More Field Testing can help us build a better Conversational AI system for farmers
e-farmerce Platform is competing in this year's UN and ESA led World Challenge in Finland!
e-farmerce Platform AI (Agriculture Intelligence) system is still in development and we hope to make a positive impact on Agriculture which is mired in crisis.
We are constantly finding new ways to increase the accuracy of our predictions and build better conversational AI system for farmers and improve the way we visualize the data using NASA WorldWind.