4,857 is the number of satellites orbiting the earth. 1,980 of which are active serving communications, observation, navigation, and science. These satellites in consolidation provide access to the earth surface like never before. Providing access to data which can reshape industries at equivalent scales. Combining this seemingly infinite stream of raw coverage with the pattern recognition abilities of computing networks we can extract information from remote locations which would be expensive to access by traditional means.
Financing in the agricultural sector is challenging especially with limited information. Lack of information to confidently finance remote locations deters growth in those areas which lead to a vicious cycle. Providing financial institutions with reliable data enables them to grant loans to farmers with confidence. These investments can be monitored over time and predictions can be made based on historical metrics such as crop yield. As clients, farmers can actively monitor and receive regular reports on their fields to maximize their harvest.
The problem can be broadly segmented into three pipelines — data collection, preprocessing and prediction. By leveraging the open source access to the Sentinel-2 satellite through the Copernicus portal. We focus on addressing the problem of preprocessing the data which enables reliable prediction. The two major issues that we categorize as preprocessing steps are described as follows,
Problem 1: Noise Removal
Access to data does not imply ready-to-use data, we have to adapt the data to our solution method. Satellites observe the Earth’s surface but succumb to the obstruction of clouds. The resulting satellite images are partial observations which are noisy due to the shadows of the clouds on the visible surface.
This problem is more specific — find and remove clouds and their shadows on satellite images. Formulated as a pattern recognition problem, we use deep learning on a small labelled set of satellite images to train a model to remove clouds and accompanying shadows. The link below is a detailed technical report of our solution to this problem.
Nimbus — Cloud Segmentation using Deep Learning for Agriculture.
Removing clouds and shadows using deep learning on open source satellite images for better information extraction.
Problem 2: Plot Detection
When our observation of the land in the satellite images is reliable and free of obstruction, we face the second problem of identifying objects. In an image, all points are seemingly independent colours and not objects such as building, tractor, farm or road. For simplicity, it is preferable to track metrics for individual objects.
Reusing the noise removal model trained with plot detection annotations generated sub-optimal but convincing results for this problem. The details of this approach are discussed in the following article,
Credit Scoring for financial institutions, portfolio optimization for investment companies, competitor analysis for strategic management, autonomous agents for field management and predictive modelling are only some of the benefactors of this technology. How A.I. will drive innovation in the 21st century is only limited by human creativity.
Artificial Intelligence has changed the way how we solve problems. In every industry from games to finances, in homes and public places, whether for business or leisure — A.I. is altering the nature of solutions. This change is not abrupt or random, the transition is gradual and steady. Let us allow technology to do what they were designed for — solve problems.
To most people this is just dirt. To a farmer this is potential.
Source: Information on Satellites orbiting Earth — https://www.pixalytics.com/sats-orbiting-the-earth-2018/