Multi-sensor imagery rectification and registration for herbicide testing

The use of multi-spectral sensors has been focused on several agricultural tasks, yet it is necessary to further assess this approach to achieve sufficient precision to carry out adequately these. Metric information from these images is traditionally derived by photogrammetric techniques, but with a major limitation: photographed objects should be static while the photographs are being taken, but plants are generally in movement because of wind and this causes the photogrammetric process to be unable to generate the necessary information to make any metric measurement. To bypass this, metric information can be derived via rectification, using only one photograph. This work aims to develop a band co-registration method with agricultural purposes, based on rectified images taken from different sensors usually mounted on UAVs or terrestrial vehicles, studying its accuracy in a quantitative way. All multispectral information co-registered in a precise way will allow the calculation or development of new radiometric and even geometric indices that will help to improve efficiency in many tasks related to agriculture. Images taken from a multi-spectral (green, near infra-red, red and red edge) and a thermal camera were used to apply the developed methodology. First, a digital elevation model describing the displacement produced by distortion due to the sensor lens was obtained and applied to each of the studied pictures to correct this distortion. Then, distortion due to conic perspective present in the photographs was corrected, taking into account the homology relationship between the photographed object and the picture. To carry out these tasks, several computers programs were developed. Subsequently, the edges of the five bands corresponding to 250 plants were digitalised and their areas were measured. Furthermore, the intersection of the five bands of each plant was calculated, and an index (AI) indicating the fraction of the area of each band, which was out of the common area edge of the five bands, was calculated for each plant. The average value of this index for each band ranged from 0.22 to 0.24, with no statistically significant differences between them, indicating a high accuracy of the proposed methodology.

Automated summary

This study developed a band co-registration method based on rectified images from different sensors, aiming to achieve precise alignment of multispectral information for agricultural purposes. The quantitative analysis showed high accuracy of the proposed methodology in improving efficiency for various agricultural tasks.