How neural networks make GIS-supported crop production more efficient

by Michael Marz 

High-tech has already arrived in agriculture. Modern agricultural machinery and geographic information systems make precision farming possible. An important aspect in crop production is the basic fertilization with nutrients and additives. The pH value can already be determined in a small scale and during the crossing of a tractor with specific sensors. For essential macronutrients like phosphorus, on-site soil sampling as well as complex and cost-intensive determination of the content in the laboratory are still necessary. Phosphorus content depends on chemical and physical soil conditions. If these conditions are known, the content of phosphorus can be estimated. This paper provides extracts from a research project on how neural networks help to estimate local phosphorous content for fertilization by learning those nonlinear multilateral relationships in the soil.

From aerial images to maps with Deep Learning

by Sebastian Gerke 

With the increasing amount of aerial imagery – regardless if captured from aircrafts or UAV – available, automatic analysis of this data becomes more and more important to make use of this data. TerraLoupe uses deep neural networks to automatically detect objects of interest for various industries such as automotive, insurance or energy to help their clients getting deeper insights for their business critical applications. This includes high-definition maps for autonomous driving, risk assessment for insurances and solar potential analysis.

Natural language understanding in web mapping applications

by Matthias Stein  , Martin Wilden  ,

“Will speech soon do away with typing” – This question has recently been raised by a large German Newspaper. Lately, more and more speech assistants like Google Assistant, Amazone’s Alexa, Microsoft Cortana or Apple’s Siri are reaching the market and all owing users to control many parts of their life by voice input. This Technology is called natural language and understanding (NLU) and may also be used to control web mapping technology. Imagine that you could say “show me all schools in Berlin” instead of using a selection tool to select those features.