Crop Identification Using Remote Sensing Technology

Abstract

Crop identification and classification plays vital role in agricultural monitoring and use of resources efficiently to develop sustainable systems and provide high productivity. However, it remains a challenge to monitor crops and their productivity due to lack of accurate crop data and system to properly identify crops with no or less effort. The present study includes digit dataset preparation tool and crop identification method for all major crop types but system is tested on wheat crop so far because in Pakistan, wheat is important both agriculturally and economically. The main goal of this study is to provide a tool for crop data collection, tool for refining the data as per our need, dataset preparation for machine learning model using satel lite raw bands and finally use models to classify crops. Android mobile application is developed and published on Play Store to collect data for different crops which includes co-ordinates, sow data, harvest date, area and crop type. Web-based application is de veloped to visualize collected data under android application. Web app also provides options to manually adjust the farm boundaries and parameters collected. Sentinel Hub official process API is used to get Sentinel-2 l1c and l2a satellite bands data to calculate different indices like NDVI, NDWI, EVI and True Color to refine the data collected for wheat crop considering crop cycle of wheat in Pakistan. After refining data with the help of process API and web-based app, input data for machine learning model is pre pared using raw values of all 12 bands of Sentinel-2-l2a. Random forest and Light GBM models are trained and testing on the above dataset with the highest 99.47% training accuracy and 88.9% testing accuracy.