Visual Recognition in Agriculture

Different Persona using Same Data  : The user  ( farm owner. Agri worker , Tomato Sales shop, Tomato buyer ) accesses the website to get to know the tomato field. Purpose of each person might be different, though that is the same for all.  Web browsers directly access the MQTT service. Application’s data pump polls the agriculture field every minute looking for new data. Data is pushed to MQTT service. Archiver is subscribed to the MQTT service and sends all new data to Influx DB. On any new data it computes current state of tomato crop in a given field, and publishes both it and the recent time series to the MQTT service

Agriculture sector brings up major challenges in  handling workflow to monitor the health of growing crops.   Most workers look at crop growth on a daily basis and take a decision on  “to apply pesticide or not”.  If there is a delay in applying pesticide then the crop will not yield good harvest.  Worker cost per day increased and also not many young people have the inclination to take up work on a farm on a daily wage basis.  Added to this, there is a need to have capital to train these workforce and deploy in the field.  All these added up to the level in which farmland owners get nervous to go in for short term crops such as potato, tomato, wheat etc.

IBM Watson Studio based Visual Recognition service  is used to build an application that can be a Digital Assistant to Agriculture workforce in the Agriculture industry.  In case, if this is expected to work locally then it is required to have local deployment of infrastructure (PowerAI) for Visual Intelligence Service.  Leveraging automation to manage investments from enhance agriculture workers productivity to  to market need based innovation on agriculture production.

Identifying crop disease and acting on insights faster with machine learning optimization. This will lead sustained output from harvest and also provide relief to agriculture farm owners to manage cash flow well.

Deploy AI-based applications in agriculture farms in large scale by  prototype to using on premise PowerAI enabled cognitive computing infrastructure. In this direction, IBM cognitive computing infrastructure appears to be best fit to deliver high performance computing requirements at low cost.

Case Study: Monitor Tomato Farm  “Customization ready Visual Recognition Micro service”

Nalla Paaru  is a Mobile Application in Android Phone. Tomato packing line worker use Visual Intelligence Micro Web Service to become part of workflow to monitor and deliver good quality Tomato. 

Visual Recognition

During monitoring, workers can be efficient  by using NallaPaaru  Digital Assistant to check the quality of Tomato. Application NallaPaaru provides Mobility to workers and workers can carry NallaPaaru in their Android Smartphone and use it effectively. 

Cost of per diagnosis is a critical parameter and complexity of workflow to perform diagnosis is another parameter.  Nallapaaru is addressing both these parameters by using IBM Watson IoT platform to reduce complexity in workflow and Visual Recognition platform to reduce cost per diagnosis. Innovation in creating Optimal yet Robust models by using  Deep learning convolutional  Neural  Networks had led to Low cost Nallapaaru. For example,  Agri workers start diagnosis  work and get results within 2 to 3 minutes by using Smartphone apps with few clicks ( sub  5 clicks). And also cost per diagnosis is  5 Rs.  Workflow complexity for diagnosis is removed and this is brought down to few clicks in Smart Phone Application

Plant Disease Detection Using Convolutional Neural Networks with PyTorch