ML Based Remote Sensing Mobile and Web Solution for Crop Area Estimation by Farmers and Field Staff

Due to climate change, frequent occurrences of extreme weather events are causing severe damage to crops. Insurance agencies need to quickly and accurately assess area, health, and ownership of crops, before and after these events, to compensate the affected farmers. However, such a quick assessment is difficult if it has to be undertaken through field visits; field visits also affect objectivity of the process. Hence, there is a need for a remote sensing solution. The European Union has successfully used such an approach to release subsidies to farmers by: 1) getting them to upload their cultivation details using a custom mobile application, along with geotagged and time stamped photographs, 2) getting these details verified at a centralised facility by a team, 3) followed by occasional farm visits. Above approach will need to be modified and adopted to the Indian context since farm sizes here are small and crop variety is large. One such modification proposed in the current work is to get farmers and field staff to undertake the verification process. Such a participatory approach is known to build confidence and trust among stakeholders. To empower them for this purpose, an ML based mobile and web remote sensing solution with easy-to-use interfaces is developed. It contains two mobile applications: 1) one with a local language interface for use by farmers to upload cultivation details, 2) another for use by field staff to verify these details using ML based remote sensing technology, with Google Earth Engine API in the back end to acquire satellite imagery and carry out supervised classification. Field trials using 10m Sentinel-2 satellite imagery and classification and regression tree algorithms were carried out in a south Indian village. Overall accuracy calculated using the confusion matrix was observed to be 86% reducing the need for field visits substantially.