Customized Shape Detection Algorithms for Radiometric Calibration of Multispectral Imagers for Precision Agriculture Applications

Dual panel relative radiometric calibration is an important tool for multispectral imagers mounted on UAV's for small farm precision agriculture. A customized dual panel detection technique integrated into the multispectral calibration routine is developed in this work. Otsu segmentation was the most precise method to find square reflectance panels with a controlled background. Canny edge detection proved less noisy than Laplacian of Gaussian filters and more robust to environmental changes than Otsu's method. More post processing on the images was required inside of edge detection algorithms, as the zero crossing edge detection methods amplified noise inside the image. Both hole filling algorithms and morphological filters were employed to reduce the noise. Morphological erosion filters caused under segmented images. This resulted in the desired low false positive rates, and negative volume similarity metrics for the regions of interest. A fast and reliable dual reflection panel detection technique was implemented for radiometric calibration for small farm monitoring where time is of the essence.