Crop insurance Automation
This PoC demonstrates the classification of crop insurance claims using scientific approaches and AI. It leverages geospatial imagery, machine learning, and weather analytics to automate and accelerate claims assessment, reduce manual errors, and detect fraud.
Target Users
- Claims Adjusters
- Claims Analysts
- Claims Managers
Business Benefits
- Enhanced operational efficiency
- Reduced processing time
- Improved accuracy and consistency
- Better resource allocation
How It Works
1. Image analysis: The system uses a combination of CNN-based binary classification and Vision Transformer for image analysis. It first identifies the crop type and damage type, then classifies the damage severity. The system also uses image segmentation with YOLOv8 to identify the affected areas.
2. Weather analysis: The system uses weather analytics to assess the risk of fire, drought, pest, disease, flood, wind, and storm by calculating metrics and indices to determine the likelihood of the weather event claimed.
3. Geospatial analysis: The system uses geospatial analysis to identify the location of the damage and the affected areas. It uses the Sentinel-2 satellite imagery to calculate spectral indices and other metrics to determine the severity of the damage.
Operational Impact
- Significantly reduced time to process crop damage claims
- Real-time damage assessments in under 5 minutes
- Reduced manual errors and operational costs
Customer Impact
- Faster resolution for insurer and insuree
Technical Highlights
Automating Crop Insurance Claims
This application automates the processing of crop insurance claims using predefined sample inputs.
Loading sample inputs...
Business Context
Crop insurance is essential to protecting farmers against unforeseen events such as weather damage, diseases, pests and accidents. In today's landscape, with an ever-increasing number of fraudulent claims, it has become a challenge for insurance providers to process the high volume of claims efficiently. It is extremely time-consuming and labor-intensive to validate a claim, therefore increasing operational costs. Moreover, the human error rate, due to negligence and lack of motivation for several reasons such as delayed compensation and monotonous work, is high.
Problem Statement
There is a dire need for efficient, fast and reliable solutions to assist adjusters in crop insurance claims processing. Leveraging AI and advanced analytics using weather statistics and geospatial imagery, we aim to develop a solution that can accelerate claims assessment, reduce manual errors and detect fraudulent activities.
Impact and Importance
An automated system as such could significantly reduce the time required to process crop damage claims, with real-time damage assessments generated in under 5 minutes. This represents a substantial improvement over traditional manual assessment methods, which can take days or even weeks, enhancing the quality of work and time management for adjusters and faster resolution for the insurer and the insuree.
Developer Setup
To set up and run this PoC locally, follow these steps:
- Ensure you have Python 3.9+ on your system.
- Clone the repository containing the PoC code. Navigate to the folder and install the dependencies:
pip install -r requirements.txt - Navigate to /src, and using .envsample file, set up .env file to store all required API keys and secrets.
- Ensure that at least 16GB of GPU resources are available on your system. You can check this by running the following command on terminal:
nvidia-smi --query-gpu=memory.used,memory.total --format=csv,nounits - Start Quart service by executing:
python main.py - Update api calls in nextJS code to use the locally hosted Quart service endpoint.
How to Use This PoC
Pick any sample input you want and hit 'Run'.
What happens behind the scenes?
GeoSpatial Imagery:
We use location and date of incident to process the following using Sentinel-2 Satellite
- NDVI: Normalized Difference Vegetation Index, measures vegetation health by comparing red and near-infrared light absorption.
- EVI: Enhanced Vegetation Index, similar to NDVI but reduces atmospheric interference using blue light data.
- OSAVI: Optimized Soil-Adjusted Vegetation Index, an improved NDVI that accounts for soil brightness.
- NDMI: Normalized Difference Moisture Index, measures vegetation water content using near-infrared and shortwave-infrared bands.
- NBR: Normalized Burn Ratio, identifies burned areas using near-infrared and shortwave-infrared bands.
Image Classification:
We apply advanced machine learning techniques to classify crop damage based on images uploaded by farmers
- CNN-based Binary Classification: Convolutional Neural Networks (CNN) are used to classify crops as either healthy or damaged.
- Vision Transformer (Llama): A transformer-based model is used for image understanding and to assess the likeliness of crop damage.
- Image Segmentation with YOLOv8: Ultralytics YOLOv8 is used for image segmentation, assisting the CNN model by segmenting crop leaves and aggregating predictions for multiple leaves to improve accuracy.
Weather Analytics:
We calculate several risk indices based on weather data based on location and date of incident
- Fire Risk Index (FRI): Assesses fire risk by evaluating temperature, precipitation, humidity, wind speed, and soil moisture.
- Drought Stress Index (DSI): Evaluates drought conditions by considering precipitation, evapotranspiration, soil moisture, and temperature.
- Pest Risk Index (PRI): Quantifies pest risk using temperature and humidity data relevant to pathogen-friendly conditions.
- Disease Risk Index (DRI): Assesses the likelihood of disease based on temperature, humidity, and soil moisture.
- Flood Risk Index (FLRI): Measures flood risk using cumulative precipitation and soil moisture data.
- Wind Damage Index (WDI): Determines wind damage risk using wind gust speed and frequency of high wind days.
- Storm Risk Index (SRI): Evaluates storm risk by considering factors like dew point spread, relative humidity, wind speed, cloud coverage, and precipitation.