Main Content

Aerospace Classification of crops and land types

Objective

Reconnaissance of different types of terrain by aerial photography:

Classification of crops and land types

Comments

  • The different types of terrain can be seen by the naked eye in different colours. This is the kind of information that the system should be able to detect and segment, free from noise.
  • In particular, the different types of crops, vegetation, urban areas, water, etc. should be clearly identifiable.

The problem

The following figure shows an enlargement of the upper central part of the image: here you can see the different regions that, on first sight, seem almost homogenous, with well-defined contours that are easy for the naked eye to identify.

Classification of crops and land types

However, if we enlarge the scale to pixel level, the true problem facing computer detection systems appears:

Classification of crops and land types

The regions which, seen in a global context, appear homogenous and divisible, have turned in a local or pixel context into clouds of different coloured dots that are easily confused between each other. The contours have disappeared and in the same region colours or shapes that could be attributable to other objects have appeared. How can we identify contours in these circumstances?

Classifying pixels based on means and variances is totally impractical, given that the large level of noise in the image would invalidate the results. Furthermore, a classification does not give semantic information on the objects identified, but just statistical information, which is subject to a considerable margin of error.

Note: pixel "classification" is the usual system used for tagging aerial images. Given a series of known means and variances that are used as a model for the different crops and a distance measurement that provides the similarity between a given pixel and these models, each pixel in the image is tagged in accordance with the model that has the least distance from that pixel.

The solution

The following figures show the results obtained from segmentation tests using Imagiam’s automatic shape recognition algorithms.

  • The image on the left-hand side represents areas of the original image and the one on the right shows the results corresponding to the analysis of this zone.
  • Each region, crop or plot identified appears in a different colour.
  • All these results have been obtained without any kind of human intervention. In other words, the image processing is totally automatic. The identification system takes the left-hand image as input and generates the right-hand image as output cleanly and instantly.

Conclusions

  • There is a semantic equivalence between the objects identified and the content of the original image. In other words, the regions identified effectively correspond to the actual content of the image.
  • The algorithms used are robust against the noise of the signal and provide well-defined contours, even in areas of low definition.
  • The segmentation system divides the image correctly. In other words, the algorithm acts as an intelligent recognition phase that allows an image seen as a "rectangular matrix of pixels" to pass on to another seen as "a set of objects that make up the image".
  • These objects can be handled and managed at a level of understanding similar to the interpretation of the human visual system, opening up the road to high-level systems that search and analyse images just as the human eye and brain would.

Prev - Next >>

Lenticular

Company

About Imagiam

Imagiam provides a scalable multiplatform solution for creating and printing lenticular images with 3D effect. Imagiam is a leading and recognized lenticular solution that is used by the major offset printer manufacturers –Heidelberg, KBA-MetroPrint, Presstek– and large format printer manufacturers –FujiFilm Europe, Océ, Gandinnovations, HP– in many of their Demo Centers around the world. Imagiam has satisfied customers in more than 45 countries. The solution is easy to use and is considered the fastest in the market.