Artificial Intelligence System can now help in land use classification

Land Classification plays an important role in providing information about land cover, land usage and how a particular barren patch of land can be utilized for crop production. Land use classification is a complex process for various countries.

 

In Switzerland it is difficult to track land use classification as the aerial photos of land are manually classified to ascertain which part of land can be used for specific crop cultivation. It is even more difficult to assess soil permeability and soil type. There are 40 different categories of land in Switzerland in which the photos are classified. The classification process is done by human beings.

 

The land survey reports in Switzerland are published after every six years. To simplify the lengthy classification process Valerie Zermatten has developed a machine learning algorithm. The algorithm recognizes different categories of land like railways, sports fields, water purification centres, campgrounds, sportsfields, airports, rivers and dams.

 

(Must Check: How has Artificial Intelligence Changed our Daily Lives)

 

The algorithm was fed with several images related to different parts of land like school buildings and power stations. The algorithm incorporates infrared images and colour photographs. It was successful in the identification of green spaces and the areas that were devoid of vegetation and buildings. This artificial intelligence program can be used for more detailed land classification in future.

 

The AI program was tested on a 600 kilometres square stretch of land in Valais(Switzerland)  from Chamoson to Sierre.  In total the algorithm was trained with 60,00 aerial photographs from which 3,000 photographs were related to green spaces like vineyards. The photographs were taken in 2020.

 

As reported by Tech Xplore Valerie Zermatten, Masters Student in Environmental Engineering and the developer of AI systems, said,”I really enjoyed starting from scratch and developing everything myself. I also wanted to work on a Master's project that had a useful real-world application."

 

The AI code of the program is now available on an open source platform. There were limited photographs available to train the AI system therefore the algorithm was programmed to assign specific weightings to different kinds of images until the margin of error is negligible. 

 

There are various problems related to farming in Switzerland. There are very small fields which make it impossible to maintain a rotational farming method. There are natural obstacles like difficult terrain and harsh climatic conditions which make it difficult to grow seasonal fruits and vegetables. Good agricultural land is extremely rare  in the scenic valleys of Switzerland.