Non-Invasive Fungicide Inspection Methodology Using Multispectral Imaging

Abstract

The purpose of the current article is to show the development of a non-invasive multispectral imaging inspection methodology for the detection of fungicides on strawberry hulls. High quality images were captured using a Micasense RedEdge multispectral camera that captures images with a resolution of 1280 x 960 pixels in its five different spectral bands. The image acquisition process was performed in an environment controlling the stability of the light spectrum, the luminous intensity, the direction of the incident light and the focusing distance. To facilitate image preprocessing, a black background was used to optimize the segmentation process of the image of interest, facilitating the isolation and cropping of the object of interest to reduce the dimensionality of the images. The contamination of the samples with the selected fungicide was performed in the laboratory at 5 and 10 parts per million (ppm) to facilitate the validation of the classification results. The study included the combination of different spectral bands generating vegetation indices that allowed the correct classification between contaminated fruits using deep neural networks. Through the research it was possible to establish a methodology to identify the presence of pesticides in strawberries through ppm dilutions. In addition, an initial multispectral image acquisition system was proposed that minimizes the impact of external variables. Finally, a preprocessing and classification algorithm was developed, demonstrating that the crop growth index (CGI) allows detecting the chemical on strawberry surfaces.