Sim A Field

Go to download link


In precision agriculture, crop/weed discrimination is often based on image analysis but though several algorithms using spatial information have been proposed, not any has been tested on relevant databases. A simple model that simulates virtual fields is developed to evaluate these algorithms. Virtual fields are made of crops, arranged according to agricultural practices and represented by simple patterns, and weeds that are spatially distributed using a statistical approach. It ensures a user-defined Weed Infestation Rate (WIR). Then, experimental devices using cameras are simulated with a pinhole model to create pictures of the field from a virtual camera.


We propose this program to help you evaluating your crop/weed discrimination algorithms and hope it will save you a lot of time.


The following text roughly explains the modeling process; you can find a detailed description in article [1].

1. Field modeling

A spatial crop field model is designed using a predefined spatial distribution of crop and weed plants. This model divides into two steps: (1) simulation of the crop field where weed plants are described by spatial stochastic models (uniform or cluster processes) and (2) construction of a virtual photograph of the crop field according to the extrinsic and intrinsic parameters of the virtual CCD sensor.


The goal of the model is to test the efficiency of crop/weed discrimination algorithms in crop field subjected to different Weed Infestation Rates.

1.1. Crop model

Crop sowing complies with reality: lines are equally spaced with an inter-row width depending on the crop type (16-18cm for wheat, 45 cm for sunflower, etc.). Two different sowing schemes are proposed: an “in-line” one for cereals and a periodical one for sunflower or maize. Periodical sowing introduces the notion of intra-row frequency, which is defined as the distance between crop plants in the row. For each crop, three different patterns and sizes are available (with three different orientations) to mimic the unpredictable nature of plant growth. The presence of crop is controlled by a stochastic variable to reproduce growth issues.

1.2. Weed model

Weeds, which are also described by different patterns, are placed in the field according to spatial distributions. As for crop plants, they can be presented at different growth stages and their size varies as a function of a stochastic variable. In crop field, two weed spatial distributions are usually observed: uniform and aggregate. As it is unlikely that weeds follow only one distribution, a mixture of both is also proposed. The Weed Infestation Rate (WIR) is specified by the user and allows the evaluation of crop/weed discrimination algorithms applied to different infestation rate.


Now that we have modeled a complete field, we need to simulate the picture acquisition.


Fig. 1: a) Wheat picture: inter-row width = 16cm, camera’s height = 1.20m, WIR = 20% with a mixture of both distributions. b) Sunflower picture: inter-row width = 45cm, intra-row frequency = 10cm, camera’s height = 5m and WIR = 10% with a punctual distribution.
Wheat and sunflower fields

2. Virtual photographs: world to camera transformation

To create an adaptive model that simulates picture shot from a remote sensor embedded in a tractor, a drone or any other experimental platform, a simple mathematical projective model is used, the pinhole model that maps a 3D world point into picture coordinates. It requires two kinds of input parameters: the intrinsic parameters, which define the camera characteristics (CCD size, focal length, distortion), and the extrinsic parameters, which define the camera location in a 3D space (rotation and translation).


As an example, Fig. 2 presents two different pictures of two different crops (i.e. wheat and sunflower) issued from two different virtual camera configurations.


Fig. 2: a) Wheat picture: inter-row width = 16cm, camera's height = 1.20m, pitch angle = 65°, roll angle = 0° and WIR = 20% with a mixture of both distributions. b) Sunflower picture: inter-row width = 45cm, intra-row frequency = 10cm, camera's height = 5m, pitch angle = 0°, roll angle = 20° and WIR = 10% with a punctual distribution.
Wheat and sunflower fields pictures

3. Model validation

The model had to be validated to ensure that modeled pictures can be used in place of real ones.


The degree of spatial similarity between the real picture and its homologous virtual one was estimated for both crop and weed patterns from spatial descriptors at different length scale: from local information to global information. Obtained results confirm the crop and weed field modeling goodness to reproduce a real field.


The world to camera transform had also been validated by comparing a calibrated checkerboard with its virtual reproduction.

4. Conclusion

Testing discrimination algorithms in various configurations is now possible with a large range of parameters:

-     crop type and its corresponding inter-row width (e.g. 16cm for wheat, 45cm for sunflower),

-     weed spatial distribution (uniform, aggregate and a mixture of both),

-     Weed Infestation Rate (WIR),

-     camera's intrinsic (focal length, CCD size) and extrinsic (3D location and orientation) parameters.


Based on this knowledge we are able to evaluate the accuracy of any spatial crop/weed discrimination algorithms by a simple comparison between initial and detected crop and weed location.

5. Bibliography

The following articles present the model and its use to perform algorithms evaluation.


[1] Jones G., Gée Ch.,Villette S., Truchetet F., 2010. Validation of a crop field modeling to simulate agronomic images. Optics Express, Vol 18(10), p. 10694-10703. doi:10.1364/OE.18.010694. Free pdf download

[2] Jones G., Gée Ch., Truchetet F., 2009.  Assessment of an inter-row Weed Infestation Rate on simulated agronomic images by image processing. Computers and Electronics in Agriculture, Vol 67 p.43-50.

[3] Jones G., Gée Ch., Truchetet F., 2009. Modelling agronomic images for weed detection. Application to the comparison of crop/weed discrimination algorithm performances. Precision Agriculture Vol 10 (1) p.1-15.


Complete bibliography





To Top