With more satellites orbiting the earth, monitoring of fields using satellite data has become easier and ubiquitous. Frequent observations of a field can provide vital cues about field health and management practices. However, farm analytical statistics derived from such datasets often need modification to create practical applications. This paper focuses on the detection and removal of field machinery track line pixels to reduce their effect on satellite-based agronomic recommendation and product development.

Two methodologies were tested for detection purposes using medium spatial resolution Sentinel-2 images. In the first method, the image was passed through various filters and mathematical operators to detect the track lines. This method is quick and requires low computational resources. However, the output is influenced by infield differences and other infield objects. Additionally, it is input resolution dependent thus limiting the application to medium resolution satellite image sources such as Sentinel-2. In the second method, the content of a sliding window is Fourier transformed into the frequency domain. Signals within a given frequency range are examined for improved accuracy and precision of the detected tracks. This method is most suited for precise detection of the machinery tracks despite the high computational demand. Additionally, this method can handle various resolutions and can therefore be used for a wide spectrum of remote sensing data, for example drone or satellite imagery.

Once the machinery tracks are detected, the data on the detected pixels can be modified using neighboring pixels. Three neighborhood filters, namely mean, median and weighted mean, were tested. The median filter was found to produce the best result in removing the track line effect and allowed field statistics that are free from the impact of the track lines present in the images.